Technology

As part of Best Buy's Black Friday in July sale, the price of the Apple MacBook Air 13-inch (M5) has dropped to $1199 (was $1299).

Powered by Apple's M5 chip, it combines a 10-core CPU, 16GB of memory, and a 512GB SSD in a lightweight design. For my money, with the extra RAM and storage this makes it to model to choose if you're buying a MacBook Air.

It's an excellent choice for students, professionals, and anyone looking for a premium everyday laptop that's lightweight and loaded with long-term value - especially given Apple recently announced price increases across all its laptops and tablets.

Today's best MacBook Air deal

13.6-inch 2560 x 1664 display, Apple M5 processor with 10-core CPU and 8-core GPU, 16GB memory, 512GB SSD, two Thunderbolt 4 ports, backlit keyboard.View Deal

Should you buy it?

✅ Buy this deal if...

You want a lightweight premium laptop with strong everyday performance, and enough memory and storage for years of work, study, and creative tasks. Battery life impressed us during testing, lasting around 15.5 hours of web use and comfortably getting through a full working day on a single charge.

❌ Skip this deal if...

You need a larger display, more ports than Thunderbolt 4 provides, or dedicated graphics for demanding professional workloads and modern PC gaming. In that case, check out the MacBook Pro instead.

Why we recommend it

In his review, our laptop expert Lance said the latest MacBook Air was "the best ultraportable I've ever used". He noted it delivers an outstanding balance of performance, battery life, and portability, handling everything from everyday productivity to photo editing, 8K video editing, and even gaming without feeling out of its depth.

Despite its slim, fanless design, he found it surprisingly difficult to push the M5 chip to its limits, even with demanding creative applications and heavy multitasking.

The 10-core CPU, 8-core GPU, and 16GB of memory provide plenty of performance for work, creative projects, and everyday computing, while the upgraded 512GB SSD delivers faster storage than previous entry-level MacBook Air models, helping applications launch quickly and improving file transfers.

At just 2.7 pounds, it's easy to carry every day, while the premium aluminum chassis feels exceptionally solid despite its lightweight design.

Price Context & Historical Value

Although the $100 discount isn't huge, deals on Apple's newest MacBook Air models are relatively rare - especially with Apple recently hiking prices across all its devices. With the upgraded 512GB base storage and M5 processor, Apple's latest ultraportable is a must-have.

The Catch: What to know before you buy

This configuration offers excellent everyday performance, but it isn't designed for intensive 3D rendering, high-end gaming, or workloads that benefit from dedicated graphics.

Connectivity is also limited to two Thunderbolt 4 ports, so you may need a hub or dock if you regularly connect multiple wired accessories.

• Shop all MacBooks at Best Buy

• Software support for the Pixel 6 and Pixel 6 Pro ends soon
• The Android QPR1 will be the final update for these phones
• They were first launched by Google back in October 2021

If you're rocking a Google Pixel 6 or Pixel 6 Pro, launched in October 2021, then your five years of software updates are coming to an end: Google has confirmed that these devices aren't going to get access to the latest Android QPR2 Beta 1.

"Pixel 6 and 6 Pro users will not be getting this OTA since the device will reach End of Life and Android 17 QPR1 beta series will be the last for these devices," says Google in the release notes of the latest beta, via Android Police.

That means no more beta updates for the Pixel 6 and 6 Pro, with the final stable software release officially Android 17 QPR1, set to be part of the September Pixel Drop. That should land just before the five-year anniversary of these phones in October 2026.

As with every end-of-life situation like this, your handset isn't suddenly going to stop working. However, it will no longer get new features or security updates — the more time passes, the more vulnerable it will be to bugs and malware attacks.

Apps may start slowing down and misbehaving, and the recommended course of action is to upgrade as soon as possible. Google is all set to unveil the new Pixel 11 series on Wednesday, August 12, so maybe you could treat yourself to one of those handsets.

There's life in the old phone yet

The Pixel 6 Pro (Image credit: TechRadar)

You've got options for what to do with your old phone. Your network operator may recycle it for you and even give you some cash back (check the website for details), but if not, you can find numerous third-party services that will — just run a web search for them.

You can also repurpose your old Pixel 6 or Pixel 6 Pro and turn it into something else. These ideas will all work without needing regular Android software updates:

• Dedicated e-reader — get the Kindle app (or something similar) up on your old phone and you can keep it exclusively for getting through your e-books.
• Car dashcam — get your old phone mounted on your car dashboard and plugged in, install an app such as Droid Dashcam, and you've got yourself a dashcam.
• Alternative webcam — sticking with cameras, your old phone probably offers a higher resolution and better angles than your current webcam. Here's how to set it up.
• Portable power bank — both the Pixel 6 and 6 Pro can work as chargers for other phones when connected via USB-C (they can also charge smaller devices wirelessly).
• Wi-Fi extender — if part of your home can't get Wi-Fi, use your old Pixel as a Wi-Fi extender for your router via the hotspot option (under Network and internet in Settings).
• PC console — with an app like Touch Portal installed, you can get a Stream Deck-like experience for gaming and streaming with your old Android phone.

There's no excuse for leaving it gathering dust in a drawer, or just throwing it in the trash (and adding to our growing e-waste problem). And you've still got a couple of months to decide what you'd like to do.

• Philips is rumored to be launching a new camera-based Hue Sync device for its TV backlights
• It removes the need for a pricey Hue Play HDMI Sync Box
• The 'Sync Screen' would be the first camera for the Philips Hue Play system

Philips Hue could be working on a new device for syncing your TV’s color palette to your backlights, giving you another option for illuminating your entertainment setup

Fabian of Hue Blog attended an online presentation hosted by Philips, and spotted a new Hue Sync device with an integrated camera that looks strikingly similar to rival devices like the Nanoleaf 4D.

The new gadget could be called ‘Sync Screen’, and may be unveiled at IFA 2026 in September. We don't know how much it will cost yet, but considering Philips Hue is a premium brand, it's likely the new device will sit in a higher price bracket than its competitors.

Just like other camera-operated TV LED lights, Philips’ version mounts to the top of a TV or monitor, and captures the color movements of the movie, show, or game on the screen. This movement is then processed and synced with backlights mounted behind the display.

From the image shared by Hue Blog, it seems that the Philips Sync Screen camera will only have one lens, but will likely be a fish-eye one to capture your entire TV screen — just like those used by rival devices.

You will almost certainly still need a Philips Hue Bridge to link the lights and camera to one another, but because the system camera-operated, you won't need a Philips Hue Play HDMI Sync Box or Hue Sync TV app subscription, which could make it a more tempting proposition.

The Nanoleaf 4D TV is just one of the Philips rivals that use cameras to sync TV content with smart backlights (Image credit: Nanoleaf)

As mentioned, Philips isn’t the first smart home giant to offer TV lighting gadgets like this, which aim to make your TV and gaming experiences more immersive by extending the colors from the screen onto the surrounding walls.

In addition to the Nanoleaf 4D light strips, the Govee StarPal Pro and the more recent Govee TV Backlight 3 Pro models are all camera-operated — the latter having a dual-camera system to capture more of your TV screen or monitor.

Philips has had quite an eventful 2026 already. In March the company unveiled its new OLED TVs, which are also the first native models for Dolby Vision 2, as well as its beautiful yet expensive ceiling light that mimics natural daylight. We’re just a mere two months away from IFA 2026, and I think the best is yet to come.

The conversation about AI in space keeps arriving at the same image: a floating supercomputer processing the world's information from 400 miles up. It’s a compelling narrative, but there’s a gap between what companies are hoping to build and what is being built today.

Today, satellites run on fixed power budgets measured in watts with strict constraints. Bandwidth is scarce enough that every byte reaching the ground has to earn its place.

Those limits push the field toward architecture that looks more like a nervous system, than a data center – lightweight models running onboard that interpret sensor data in near real time and convert observations into structured events. The event reaches the ground.

The raw pixel doesn’t.

From hours to minutes

In a conventional Earth observation pipeline, a satellite captures an image, downlinks it to a ground station, ground systems process the raw data, and the result reaches whoever needs it. Best case that happens in hours, but often it’s closer to a day.

For a disaster response team managing a flood in a low-lying river delta, or a conservation authority trying to locate the origin point of a wildfire in a remote national park, that day comes at a cost.

A satellite running inference onboard changes the math. Detection happens in seconds. What gets downlinked is a position, timestamp, or risk score. The bottleneck shifts from the space segment to the ground distribution, which is a comparatively manageable problem.

The use cases where this matters most are not the visible disasters people are watching. They’re the methane leak on a pipeline with no weekly inspection schedule, an oil spill beyond the reach of coastal patrols, a wildfire that began in a remote area before anyone had reported smoke. Onboard inference turns a passive imaging asset into an early warning system.

What orbital constraints teach edge architects

The tradeoffs being resolved in orbit are an extreme version of the same constraints facing any organization deploying AI outside a well-provisioned data center.

Cloud-native AI development often has a back up plan: when the model is too large add compute; when bandwidth is constrained, increase it; when latency is a problem, move the processing closer. In orbit, none of these options exist. You build within the envelope, or the system doesn’t function.

The result is a forcing function that enterprise architects rarely encounter at the same level. Industrial IoT deployments face intermittent connectivity. Autonomous systems can’t afford round-trip latency to a central server at decision time.

The shift from 'send everything, process centrally' to 'process locally, transmit what matters' is happening across multiple industries. Space is where that shift ran without a safety net.

The bandwidth math

The data reduction numbers transmitted in real time is not just 80-90 percent. Once processing happens on the spacecraft, the reduction for the real-time layer exceeds 99 percent. This is semantic compression. The satellite sends the meaning of what it saw, not the measurement it produced.

A conventional operator downlinking hundreds of terabytes of raw imagery daily is paying bandwidth cost for data that largely contains nothing of interest. With onboard inference, what's transmitted in real time is a structured detection event: a position, a timestamp, a risk score, and perhaps a small compressed image. That is hundreds of kilobytes, not terabytes.

An operator downlinks only what warrants examination, rather than blindly dumping the full data stream.

Architecture decisions that preview what's next

A model making decisions before a human is in the loop carries different requirements than one generating recommendations for human review. Ambiguity tolerance is lower. Inference behavior needs tighter scoping. This is the same conversation that medicine and finance have been having for years.

AI-assisted diagnostics and accountability distributed across platform, model, training data, and end customer rather than concentrated in a single layer. The space industry is joining a conversation other sectors have already been having for years.

In a cloud environment, model size and computational efficiency are optimization targets. Meaning they are important, but secondary to capability. In a constrained orbital environment, they are the primary design constraint from which everything else follows. A model that cannot run within the available compute envelope is not a model that gets deployed. There is no option to add a larger instance.

A maritime patrol aircraft that previously ran random vessel inspections now works from a ranked list of targets with risk scores attached. Some alerts will be false positives which is a physical reality of any probabilistic system. But the aircraft's operational effectiveness improves substantially compared to random patrolling or no monitoring at all. The AI narrows the search.

The scaling problem is familiar

One satellite running an onboard model is a proof of concept. A constellation of hundreds running distributed inference is a different infrastructure problem as orbital AI scales.

Centralized orchestration becomes the bottleneck when constellations grow. Every decision can’t route through a ground station. Distributed inference is a requirement. Enterprise architects hit the same wall when a pilot deployment expands to thousands of edge nodes. The centralized model that worked in development becomes the thing that breaks in production.

The cloud infrastructure analogy supports this. Nobody builds a data center before launching an application. The pattern is shared infrastructure, with control at the model, mission logic, and decision layer. Those can be sovereign regardless of who owns the underlying compute.

A design principle worth carrying

It’s hard to develop constraint-based thinking in environments where adding compute is always on the table. The organizations that have built it tend to have faced conditions where it wasn’t.

The strategic advantage in edge AI over the next decade will not just be measured in the amount of compute available. It will also be measured in code deployed to the right place in the stack. Satellites are running that experiment first.

We've featured the best AI tool.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

The TobenONE 11-in-1 USB-C Docking Station is on sale for $83 (was $130) at Amazon, a saving of $47.20.

Thanks to its programmable shortcut keys, dual 4K display support, and a wide selection of ports, it will be especially appealing to productivity-focused Windows users.

This docking station supports dual 4K 60Hz displays, 100W USB-C Power Delivery, Gigabit Ethernet, three 5Gbps USB ports, SD and microSD readers, programmable shortcut controls, audio jack, and dual HDMI connectivity for Windows laptops.

Dual HDMI 4K@60Hz, programmable shortcut keys and volume knob (Windows), 100W USB-C Power Delivery input, Gigabit Ethernet, USB-C and USB-A 5Gbps ports, SD/microSD card readers, 3.5mm audioView Deal

Should you buy it?

✅ Buy this deal if...

You want a single USB-C dock that connects dual 4K monitors, wired networking, storage devices, and accessories while adding programmable shortcut controls to speed up repetitive Windows tasks.

❌ Skip this deal if...

You primarily use a Mac and need two independently extended external displays. The customizable shortcut features are also designed specifically for Windows systems.

Why we recommend it

In his review, our expert Mark said: "This is a clever concept, and for those with a video or photo editing workflow, it's genuinely useful."

Most budget USB-C docks follow a familiar formula, but this model takes a different approach by integrating four programmable shortcut keys and a multifunction rotary dial alongside its standard connectivity.

The controls can be configured via the TobenONE software to launch applications, take screenshots, control media playback, mute audio, or trigger custom key combinations.

Dual HDMI outputs support two independent 4K 60Hz displays on compatible Windows laptops with DisplayPort 1.4 MST support, making it well suited to coding, design, trading, and other multi-monitor workloads.

Gigabit Ethernet provides a reliable wired connection, while three 5Gbps USB ports, SD and microSD card readers, and a USB-C Power Delivery input supporting up to 100W charging complete a versatile desktop setup with a single cable.

Price Context & Historical Value

Amazon has reduced the TobenONE from $129.99 to $82.79, a 36% discount that will save you $47.20. This is as low as the dock has ever been sold, so if you've been planning to add a docking station to a home office or hybrid work setup, this is a great time to buy.

The Catch: What to know before you buy

The programmable buttons and smart control knob require the TobenONE software and are available only on Windows. Mac users can still use the dock's core functions, but customization is limited, and macOS mirrors content across two external HDMI displays rather than supporting two independently extended screens.

Our biggest criticism was bandwidth. Although compatible with USB4 and Thunderbolt hosts, the dock limits its USB data ports to 5Gbps, so fast external SSDs won't reach their full performance. It also ships without a power adapter, meaning you'll need to provide your own if you want laptop charging.

• Shop all docking stations at Amazon

The 2026 World Cup is already under way. Billions of eyes are on the pitch. The story that matters to anyone running a supply chain, though, is playing out in warehouses, customs terminals and distribution centers spread across three countries.

For the first time, the tournament spans three host nations: the United States, Canada and Mexico. That means millions of product lines, thousands of supplier handoffs and cross-border compliance requirements across three distinct regulatory environments, all compressed into a window with zero tolerance for error.

The operational scale of this tournament is without precedent.

Now that the group stages are live, the pressure on supply chains is real and immediate. The lessons surfacing are worth paying attention to, because they apply well beyond sport.

A live packaging stress test

Official merchandise for an event of this scale moves across multiple countries, customs jurisdictions and retail channels simultaneously. Product identification has to work at every stage of that journey, from the manufacturer's floor to the stadium vendor's shelf. The label that cleared customs in Los Angeles may face entirely different requirements in Toronto.

The deeper problem is structural. Today's supply chains still largely operate as disconnected islands. Each site, supplier, co-packer and carrier maintains its own systems and repeatedly re-enters the same product and compliance data. That fragmentation creates built-in waste at every handoff: redundant setup, duplicate records and inconsistent label versions. Under normal conditions, these inefficiencies are costly but can be masked by day-to-day operations. Under the pressure of a live global tournament, they become critical.

Demand shifts are happening in real time. A host city that reaches the knockout stages sees fan merchandise demand surge overnight. Supply chains built on static, batch-processed labelling data are finding they cannot respond at that pace.

The speed of these shifts can be surprisingly tangible. In Atlanta, shortly after the Spain-Cape Verde match, I walked through the airport and was struck by how many people were wearing Cape Verde jerseys. In the space of a few hours, merchandise that had been relatively low-profile had become highly visible, underscoring how quickly demand signals can emerge and spread during a global event.

From labels to live data

What the World Cup is making visible in real time is a shift that has been under way for several years. Product identification is no longer a print-and-forget exercise. It is a live data problem.

The industry is moving from fragmented, internal systems to connected, multi-partner ecosystems. The organizations managing the tournament's supply chain most effectively are those that have made this shift: rather than each stakeholder operating in isolation and recreating the same product and compliance data from scratch, they are working within a shared, real-time environment where information flows seamlessly across systems, suppliers, customers and geographies.

The benefits of this approach are measurable. Organizations that can operate with real-time visibility and trusted data across their extended value chain can reduce delays, prevent errors at source and respond faster when disruption hits. In sectors where production downtime can exceed $1-2 million per hour, that responsiveness is not a nice-to-have but operationally critical.

The cost of disconnected systems

The consequences of siloed product data are well understood in theory. A tournament of this scale is making them visible in practice.

When product data does not flow seamlessly across sites and trading partners, the failure surfaces in predictable ways. Rejected shipments at customs. Compliance failures that stall distribution. Production downtime while teams manually reconcile data across systems. Against the backdrop of a global event with fixed deadlines, those failures are not recoverable.

The organizations absorbing those costs right now are those still operating inside the organization perimeter - managing product identification as an internal function rather than a network-level capability. The distinction matters. Supply chain resilience increasingly depends on the ability to coordinate accurate product data across every site, trading partner, and customer - creating a connected ecosystem in which product identity can be shared, trusted, and acted upon seamlessly.

What happens after the final whistle

Connected, network-driven approaches to product identification are no longer a future aspiration. The World Cup is demonstrating their value in real time, at a scale most supply chains will never encounter but from which every supply chain can learn.

The direction of travel is clear. Organizations that can rapidly adapt labelling requirements across plants and partners, share trusted product data in real time and eliminate the manual rework that comes with disconnected systems will outperform those that cannot. That is as true in retail, pharma and automotive as it is in a stadium in Los Angeles.

The World Cup will be over in a matter of weeks. The infrastructure challenges it is exposing will still be there when it ends. The organizations that use this moment to address those fundamentals will be better placed for whatever high-pressure deadline comes next.

We've listed the best product information management software.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

Minisforum has two slim mini PCs discounted right now, and which one to buy comes down to how much CPU you actually need.

As part the the Black Friday in July sale, the Minisforum UM760 Slim is $600 (was $680) at Best Buy, while the beefier Minisforum UM870 Slim is $700 (was $800).

Today's top Minisforum mini PC deals

AMD Ryzen 5 7640HS (6-core/12-thread, up to 5.0GHz) with Radeon 760M graphics. 16GB RAM and a 512GB SSD, with dual M.2 slots for expansion. HDMI 2.1 and USB4 output supporting up to 8K@60Hz, Wi-Fi 6E, and Bluetooth 5.3.View Deal

AMD Ryzen 7 8745H (8-core/16-thread, up to 4.9GHz) with Radeon 780M graphics. 16GB RAM and a 512GB SSD, expandable to 96GB RAM and up to 4TB storage across dual M.2 slots. Triple display support via HDMI, DisplayPort, and USB4, dual 2.5G LAN, Wi-Fi 6E, and Bluetooth 5.3.View Deal

Should I buy it

Which to choose

✅ Choose the UM760 Slim if...

✅ Choose the UM870 if...

Workflow

Your workload is mostly everyday desktop use

You want real multi-core headroom and better integrated graphics

Connectivity

You only need to connect via HDMI and USB4

You prefer wider connectivity including dual LAN

Why we recommend these mini PC deals

Both machines share the same slim Minisforum chassis and general design philosophy, but the CPU gap between them is real rather than cosmetic.

The Ryzen 5 7640HS in the UM760 is a capable 6-core, 12-thread chip that handles everyday desktop work, browsing, and light multitasking comfortably.

The Ryzen 7 8745H in the UM870 steps up to 8 cores and 16 threads with a newer Radeon 780M GPU, and independent benchmarking has shown its multi-core performance landing in the upper third of current mini PCs, without thermal throttling under sustained load.

That extra core count and the RDNA 3-based Radeon 780M make the UM870 the meaningfully better pick for anyone doing content creation, code compiling, or casual 1080p gaming — reviewers have measured the 780M hitting 75-123 FPS in popular titles at 1080p, a tier above what the 760M in the UM760 typically manages.

Connectivity is another real difference, not just a spec-sheet footnote: the UM870 adds dual 2.5G Ethernet ports and triple display output, compared to the UM760's single display path over HDMI/USB4.

If you're setting either of these up as a home server or a multi-monitor workstation, that difference in networking and display support is worth the extra $100 on its own.

For more top picks, see our guide to the best mini PCs.

What to know before you buy

Something worth flagging on both: the blue status LED on Minisforum's UM-series machines has drawn complaints from some owners as distractingly bright in a dark room.

And if you plan to run Linux on either, the bundled Wi-Fi cards have had driver support issues reported by some reviewers — a wired Ethernet connection sidesteps that entirely.

More mini PC deals

Powered by AMD Ryzen AI 9 HX 370, this Geekom mini PC combines 32GB DDR5 memory, a 1TB SSD, WiFi 7, USB4, 8K output, and 80 TOPS AI performance for demanding workloads.

Read our full reviewView Deal

The GMKtec K16 mini PC delivers Ryzen 7 7735HS performance with 32GB LPDDR5 RAM and a 1TB SSD. Featuring OCuLink eGPU support, triple-display output, dual 2.5GbE LAN, Wi-Fi 6E, and USB4, it’s a versatile compact system for gaming and creative workloads.

Read our full reviewView Deal

• See the full Black Friday in July sale at Best Buy.

If you’ve done any shopping for marketing automation tools these days, you’ve probably noticed they all claim to be “powered by AI.” Apologies for splitting hairs, but that’s just not true.

A significant portion still rely primarily on rule-based automation, work identically to the platforms they replaced, triggering if/then workflows created by human engineers years ago. Give their systems an edge case to parse and you’ll soon see them send an inappropriate email, crash, or output some stale nonsense that wouldn’t matter to any living person.

Same label. Same old architecture. The problem? A rule bottleneck.

Traditional marketing automation relies on knowing rules ahead of time. A lead hits a score? Send an email. A prospect completed behaviors A, B, and C? Trigger sequence Y. Wait Z days, send the follow-up email.

A rules engine can execute these commands flawlessly – but it can only execute what it knows. When a situation arises that doesn’t fit the rules, what do you do? You update the rules.

Why this matters

Marketing is messy. Prospects take unpredictable journeys, trends come and go overnight, and audiences who loved your message last week don’t care about it this week. But rule-based systems can only improve when given new rules to fire. Engineers can’t possibly keep writing rules faster than the world changes.

The industry has been papering over this problem with AI buzzwords. Sprinkle some Neural Network magic on a rule engine, and suddenly you’ve got yourself an “AI platform.” Engineers who look past the updated sales brochures still find the same good old-fashioned if/then statements, patched up with trendy new nomenclature for the latest round of funding.

The difference between legacy automation and true agentic AI is that true agentic AI won’t just patch up the last generation of marketing automation tools – it will replace them. Agentic AI isn’t defined by capabilities so much as by the way decisions are made.

Rules engines ask, “what rule should fire next, given this input?” Agents ask, “what action should I take to get closer to my goal?” This is subtle but critical. Agent theory holds that the system knows its goal, its current context, and a list of available actions it can take.

Based on those three pieces of information, it can reason as to which action will bring it closer to accomplishing its overall objective. This extends far beyond executing canned responses - it’s deciding what to do.

Agentic systems maintain goals, reason over available actions, invoke tools, evaluate intermediate results, and adapt their plans as new information becomes available. The architecture is fundamentally iterative rather than purely reactive.

You know where this is going.

An agent can adapt if a campaign stops performing. It can coordinate with other agents who manage different subsets of that workflow. And it can do so without a human engineer going back into the system to rewrite the rules every time the world changes. The system manages its goals.

Why specialization matters

One important architectural decision that separates good agent implementations from the rest is specialization. Should you build one big generalist AI system to handle everything or many specialized agents, each performing their own task?

Specialization comes up often in discussions around AI, from medical doctors to Renaissance men. There is broad utility in generalization, but singular accuracy in specialization. The family doctor can handle any symptoms you throw at them. But when you need to be absolutely certain about your diagnosis, you see a specialist.

That’s because specialists aren’t smarter than the generalist – they’re just trained on narrower data. Likewise, generalist AI models aren’t going to produce great results for highly specific use cases. OpenAI’s models can write you a marketing strategy. They can craft creative assets. But they can’t produce marketing assets that:

• Fit the pixel ratio requirements of a given publisher
• Match your brand’s color palette
• Align with your target audience’s emotional affinity profile
• Incorporate mentions of trending topics from the previous day

They can’t do all of those at once, either. And you shouldn’t expect them to. For hard problems with specific solutions, you should build specialized agents (sometimes called “agent crews”) that own a narrow subset of your workflow.

One crew might specialize in strategy generation, while another focuses on creative writing. One might select publishers while another analyzes performance. Separately, these crews create atomic workflows that a generalist system would struggle to manage.

How not hosting your models affects data privacy

There’s another argument for specialized, privately hosted models that isn’t made enough: data privacy.

Whenever you use a public large language model (LLM) to write marketing copy, your data is being uploaded to someone else’s infrastructure. “We don’t use customer data for training” is easy to say but barely offers any assurance. Inputs are still being ingested, processed, stored, and handled according to what that provider’s internal policies dictate.

And those policies can change… most corporate lawyers have never looked at the data use section of public AI providers Terms of Service, let alone dissected it line-by-line.

But what about controls your organization can enforce? Do your developers scrub data for PII before generating content with an LLM? That only works if everyone in your company memorizes your data policies and uses tools responsibly. One rogue employee attaching a spreadsheet full of internal pricing to a prompt breaks your compliance.

But if the model itself is hosted privately, that’s one major source of exposure that goes away. Your data never leaves your infrastructure. There’s no ingestion point to transmit it to a third-party, no training feedback loop that will process it, and no agreement to parse about how that company will handle your data “moving forward.”

Governance is a system property

Because AI in the enterprise has reached a maturity level where governance is a legitimate concern, many teams treat it as a bulk edit at the end of AI-generated content. Have humans review and approve. That’s fine, and many teams require this today. But governance should be built into the system at a fundamental level.

Well-built agents have guardrails at every stage of the decision-making process. That means models that make predictions within set bounds. That means observability that can trace every word generated back to its origin.

That means third-party benchmarking to prove your models perform well against industry standards, not just internal testing. Governance shouldn’t just be applied to outputs – it should be inherent in the architecture.

What enterprise buyers should actually be asking about

Buying criteria for agentic AI will vary by company, but as requests for proposal accelerate to keep pace with innovation in the industry, here are a few considerations every enterprise buyer should ask about:

• Goals vs. rules - Is this system actually agentic? Or is it just automating workflows with AI tools bolted on? The first step is asking vendors point blank what their system does when it encounters data it doesn’t know how to parse. Rules engines will point to specific fallback rules that execute. Agents will talk about reassessing their goal and weighing their available actions until they decide on the next best step.
• Models and hosting - Where are the models hosted? Are they specialized and trained on domain-specific data? This answers two questions at once – vendor capability, as well as data privacy concerns.
• Long-term memory and context - Enterprise agents become dramatically more useful when they retain organizational context over time. Rather than treating every interaction as a new conversation, they can accumulate institutional knowledge, remember previous decisions, and personalize future actions while remaining within governance boundaries. Persistent memory allows agentic systems to improve continuously without requiring engineers to encode new rules after every edge case.
• Hallucination - No current LLM is immune to hallucinations. The important architectural question is how the system detects, bounds, and mitigates them before they affect downstream business processes. Specialists hallucinate less in their domain of expertise. Prediction window guardrails limit how far an AI system can go “outside the data.” Human approval gates before sending anything live catch anything that slips through.
• Governance / auditability - Can the system provide traceability for every output it generates? Is the system’s accuracy benchmarked against a third-party, or just internally verified?

The economic case for getting this right

There's an additional argument that often gets overlooked in discussions focused on capability: cost structure.

Token-based pricing from large model providers creates a fundamentally unpredictable cost model for enterprise deployments. Every question, every generation, every iteration costs tokens — and iterating toward an acceptable output for a complex campaign task can consume a significant volume of them.

Enterprise subscriptions impose usage caps that create their own operational friction. The more AI-dependent your workflows become, the more acute this pressure grows.

Organizations that own and host their own specialized models are not subject to this dynamic. There is no token meter running. The economic relationship is closer to infrastructure than to a metered service - you bear the cost of building and maintaining the system, and in return you have predictable marginal cost. For organizations at scale, that math changes substantially.

Don’t fall victim to marketing speak

AI marketing platforms will continue to flood the market with AI-sounding languages attached to rules engines. But for enterprises who truly want to deploy agentic AI, there’s a far better solution. Domain specific, privately hosted agents that don’t leave your organization exposing itself to risk.

As agentic systems mature, the organizations that differentiate between genuine autonomous architectures and AI-enhanced workflow engines will be better positioned to capture sustainable competitive advantage.

We've featured the best business intelligence platform.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

Lenovo Yoga Slim 7x Gen 11: Two-minute review

The Lenovo Yoga Slim 7x Gen 11 is an ultra-thin laptop with a strong spec that makes it more than capable for workday use.

It looks smart, thanks to its minimalist form. All sides are flat, and the Cosmic Blue colorway is subtle yet helps to distinguish it in this largely monochromatic sector.

The build quality is excellent. The body is as premium as it comes, feeling very solid but also light and smooth to the touch. In fact, it could probably lay claim to having one of the best laptop constructions around right now. Allied to that lightness is its thinness, which is quite remarkable, although the thick rear foot underneath somewhat negates this aspect, which is a shame. What’s more, these truncated dimensions mean that you get three ports, all of which are USB-C.

(Image credit: Future)

With the new Snapdragon X2 Elite chip equipped, the Yoga Slim 7x Gen 11 is a capable all-round performer. It handles multiple browser tabs and 4K streaming with ease, as well as basic productivity. While it's not necessarily aimed at gamers and lacks a dedicated GPU, it's even capable of some light gaming; it managed to run Cyberpunk 2077 in a playable state, although don't expect top-tier performance. Heat and fan noise are generated when such workloads are conducted, but neither is too disruptive.

Another advantage of the X2 Elite chip is the improved efficiency. This is evident in the battery life of the Yoga Slim 7x Gen 11, which is one of its most impressive aspects. It lasted close to a full day when I ran a movie on a continuous loop, which ranks among the best in class. It’s also quick to charge, taking about two hours to do so.

The 2.8K display in my review unit was sharp, bright, and vivid, although it was prone to showing reflections at times. For the most part, though, these were only apparent in the most unfavorable of lighting conditions.

The keyboard is fairly easy to use, thanks to the ergonomic keycaps and their spacing. However, presses are a little heavier than you might expect, so this can take some getting used to if you’ve come from a board that requires a light touch. The touchpad is large and smooth, which makes for easy navigation, while taps and clicks are responsive and provide good feedback.

The Yoga Slim 7x Gen 11 commands a reasonably high price tag, but considering its spec and quality, it makes for good value compared to some of its less premium-feeling rivals. It might lack the ports and graphical performance of other laptops, but it’s a fine choice if portability and build quality are among your top priorities.

Lenovo Yoga Slim 7x Gen 11 review: Price & availability

(Image credit: Future)

• Starts from $999.99 / £945 / AU$1,999
• Available now in one colorway
• Reasonably expensive for the spec

The Yoga Slim 7x Gen 11 starts from $999.99 / £945 / AU$1,999 and is available now in one color: Cosmic Blue. This base spec features a Snapdragon X2 Plus X2P-42-100, 16GB of RAM, 512GB of storage, and WUXGA display. This isn't exactly cheap, and costs escalate quickly as you move up the model range. The top-spec unit, which features 32GB of RAM, 1TB of storage, and a 2.8K display, costs $2,299.99 (about £1,700 / AU$3,300).

If you’re looking for another thin and light laptop, then the HP OmniBook 7 is a very strong alternative. I was very impressed with its performance when I reviewed it, and its only missteps are the port placements and slightly harsh keyboard. It's base model is slightly cheaper than the Yoga Slim 7x Gen 11's in the US, and significantly so in the UK.

There’s also the Asus Zenbook A14, which is incredibly thin and light, and has one of the best battery lifespans around. However, it lacks the build quality and display brightness of the Yoga Slim 7x Gen 11, yet it’s significantly more expensive when comparing like-for-like models.

• Value score: 4 / 5

Lenovo Yoga Slim 7x Gen 11 review: Specs

Base spec

Max spec

Price

$999.99 / £945 / AU$1,999

$2,299.99 (about £1,700 / AU$3,300)

CPU

Snapdragon X2 Plus X2P-42-100 Processor (4.04 GHz)

Snapdragon X2 Elite X2E-80-100 Processor (4.70 GHz)

Graphics

Integrated

Integrated

RAM

16GB LPDDR5X

32GB LPDDR5X

Display

14-inch WUXGA (1920 x 1200), OLED, Glare, HDR 500 True Black, 100% DCI-P3, 400 nits, 60Hz

14-inch 2.8K WQXGA+ (2880 x 1800), OLED, Glare, HDR 1000 True Black, 100% DCI-P3, 500 nits, PureSight Pro, 120Hz

Storage

512GB SSD M.2 PCIe Gen4

1TB SSD M.2 PCIe Gen4

Ports and Connectivity

3x USB-C (USB4 40Gbps); Wi-Fi 7, Bluetooth 5.4

3x USB-C (USB4 40Gbps); Wi-Fi 7, Bluetooth 5.4

Battery

70Wh

70Wh

Weight

2.6lbs (1.2kg)

2.6lbs (1.2kg)

Dimensions

12.3 x 8.7 x 0.6 inches / 312 x 221 x 14mm

12.3 x 8.7 x 0.6 inches / 312 x 221 x 14mm

Lenovo Yoga Slim 7x Gen 11 review: Design

(Image credit: Future)

• Extremely thin
• Premium materials
• Lacks ports

The minimalist design of the Yoga Slim 7x Gen 11 is appealing, while the dark Cosmic Blue colorway rescues the unit from looking dull. The rounded sides and corners help to soften its appearance. Most sides are flat, with no extraneous bulges, save for the central camera bezel that extends beyond the rest of the lid. However, the overhang this creates actually helps you to open the lid, which is a thoughtful touch.

The body is one of the best I’ve seen in a laptop. It looks and feels premium, rivaling what the best MacBooks have to offer. It’s incredibly smooth and seems very durable. There’s some flex to the base and the lid, but they’re remarkably solid when you consider just how light the whole unit is. The lid is also very stable when open, and adjustments are smooth and easy to make.

(Image credit: Future)

Just as impressive is the thinness of the Yoga Slim 7x Gen 11. It’s certainly one of the thinnest 14-inch laptops I’ve seen, although it’s a shame the bulky bar underneath, which acts as the rear foot, somewhat undermines this aspect. This is no doubt to improve airflow from the underside vent, which is more exposed than many others.

To achieve this thinness, port selection has been compromised. There are only three ports, all of which are USB-C, which means you’ll need plenty of adapters (or a hub) if you’re going to be connecting lots of peripherals. The saving grace is that all ports have 40Gbps transfer rates, support the Power Delivery standard, and can be used to charge the laptop itself. They’re also split across both sides of the unit, which improves practicality.

On the right you’ll also find the power button and a button for disabling the webcam, which might disappoint those hoping for a physical privacy shutter.

• Design score: 4 / 5

Lenovo Yoga Slim 7x Gen 11 review: Performance

(Image credit: Future)

• Capable everyday performance
• Light gaming possible
• Keys are a little heavy

Lenovo Yoga Slim 7x Gen 11 benchmarks

3DMark: Night Raid: 40,637; Fire Strike: 7,911; Steel Nomad: 1,051; Solar Bay: 21,754; Solar Bay Unlimited: 22,518; Solar Bay Extreme: 2,491; Solar Bay Extreme Unlimited: 2,526
Geekbench 6.5: Multicore: 20,362; Single-core: 3,815
Cinebench R23: Multi Core: 13,071; Cinebench R24: Single Core: 90; Multi Core: 723
Crossmark: Overall: 1,896; Productivity: 1,703; Creativity: 2,171; Responsiveness: 1,742
Passmark Overall: 7,821.2; CPU: 30,341.4; 2D Graphics: 594.2; 3D Graphics: 6,632.9; Memory: 3,429.5; Disk: 51,620.7
BlackMagicDisk: Read: 5,214MB/s; Write: 4,947MB/s
HandBrake 4K to 1080p: 63.68fps
Total War: Warhammer III: 1080p, Medium: 43fps
Total War: Warhammer III: 1800p, Ultra: 11fps
Battery Life (TechRadar movie test): 23 hours and 30 minutes

For everyday use, the Yoga Slim 7x Gen 11 is certainly fast enough. It handled many of the tasks I threw at it with aplomb, from basic productivity to 4K streaming. With the 32GB of RAM in my unit, it also handled multi-tab web browsing with ease.

No doubt a lot of this performance comes from the new Snapdragon X2 Elite chip inside the Yoga Slim 7x Gen 11. Qualcomm claims this chip is a stronger performer than Intel equivalents, and based on our testing this appears to be true. For instance, it beat the Geekbench Single Core benchmark score of the HP OmniBook 7, with its Intel Core 5 220H, by over a thousand points, and doubled its Multi Core score.

When it comes to gaming, the Yoga Slim 7x Gen 11 was less competent, but that's hardly a surprise — it's not really aimed at wooing gamers, given it doesn’t have a dedicated GPU. However, I still managed to run Cyberpunk 2077 in a playable state, albeit with ray tracing disabled and resolution scaling set to Balanced. It wasn’t exactly a smooth experience, so it's not likely to be for you if you're in the market for a dedicated gaming laptop, but if you want a productivity powerhouse with just a little casual gaming on the side, it should suffice.

As expected, the fans can be heard whirring away in such cases, but the noise isn’t too disruptive. Heat is also generated, and temperatures can get quite high in places, although thankfully these hotspots are confined to the top and underside of the base.

(Image credit: Future)

The display in my review unit was crisp, thanks to its 2880 x 1800 resolution, and the high levels of brightness ensure content is viewable in most conditions. There were times when reflections showed on screen, but only when the lighting in my environment was particularly unsuitable.

The keys in the Yoga Slim 7x Gen 11 adopt Lenovo’s trademark shape, which are very ergonomic. The generous spacing between them also makes for a forgiving typing experience. However, they’re a little heavier and travel further than you might expect, which might throw off those who prefer a light and snappy response. The touchpad is suitably large and incredibly smooth, making it conducive to cursor navigation. Taps are responsive without being overly sensitive, and clicks are easy to perform and provide plenty of feedback.

• Performance score: 4 / 5

Lenovo Yoga Slim 7x Gen 11 review: Battery life

(Image credit: Future)

The battery life of the Yoga Slim 7x Gen 11 is very impressive. No doubt that aforementioned new Snapdragon chip plays a major part here, due to its improved efficiency. It managed to last just under 24 hours in our movie playback test, putting it ahead of many other laptops in this sector. This may be partly due to that aforementioned chipset, which Snapdragon claims brings improved efficiency. Charging is a quick affair, too, as it took about two hours to fully replenish.

However, there are those that can match and even outlast it, such as the HP OmniBook 7, which managed 26 hours in the same test. And then there’s the Asus Zenbook A14, which managed a staggering 28 hours and 25 minutes.

• Battery life score: 4.5 / 5

Should I buy the Lenovo Yoga Slim 7x Gen 11?Scorecard

Attributes

Notes

Rating

Value

It seems expensive, but when you consider its design and spec, it’s reasonable.

4 / 5

Design

I couldn’t help but marvel at its thin and premium design. The port selection is disappointing, though.

4 / 5

Performance

Great for everyday tasks, and it can even handle light gaming. The display is vivid if a little reflective, and it can get hot under load.

4 / 5

Battery life

Among the best in class, it really can last a full day. It’s quick to charge, too.

4.5 / 5

Total Score

The Yoga Slim 7x Gen 11 is great if you need a thin yet powerful Windows machine that lasts all day unplugged. A few areas are less impressive, but it’s still a good value proposition.

4 / 5

Buy it if…

You want one of the thinnest laptops around
That bulky foot underneath might spoil things a little, but the Yoga Slim 7x Gen 11 is still incredibly thin.

You want a premium design
It looks premium, and the body feels exquisite and hardwearing. It’s also quite light.

Don't buy it if…

You want lots of ports
There are only three ports on board, and all are USB-C, so you'll need a hub if you have lots of connections to make.

You want light keys
Although they’re comfortable to use, the keys on the Yoga Slim 7x Gen 11 are a little heavier than on many other laptops, which might take some getting used to.

Lenovo Yoga Slim 7x Gen 11 review: Also consider

HP OmniBook 7 14-inch (2025)
Like the Yoga Slim 7x Gen 11, the OmniBook 7 is an incredibly thin and light laptop. Not only that, I was very impressed with how well it handled all kinds of workloads, and it has a superb battery life to boot. Read our full HP OmniBook 7 14-inch (2025) review.

Asus Zenbook A14
Another portable hero, the A14 is even lighter than the Yoga Slim 7x Gen 11. However, its battery life is even better; in fact, it’s the most enduring laptop I’ve ever tested, lasting a staggering 28 hours in our movie playback test. Its build isn’t as premium as the Yoga Slim 7x Gen 11’s, and its display isn’t as bright, but it’s still a capable machine. Read our full Asus Zenbook A14 review.

(Image credit: Future)How I tested the Lenovo Yoga Slim 7x Gen 11

• Tested for several days
• Used for a variety of tasks
• Plentiful laptop experience

I tested the Yoga Slim 7x Gen 11 for several days, during which time I used it for a variety of tasks, from productivity and browsing to streaming and gaming.

I also ran our series of benchmark tests, designed to assess every aspect of a laptop's performance. I also ran a movie on a continuous loop to test the battery life.

I've been using laptops for decades, and have reviewed a large number of them, from small budget models to large gaming machines.

• First reviewed: July 2026
• Read more about how we test

Data center design has been shaped by a familiar set of priorities for years: keep systems available, resilient and predictable in any condition. Just like the electrical grid that powers these sites, they have been engineered to provide a highly consistent service regardless of what happens, even when individual components fail.

This has meant operators build layers of redundancy into power, cooling and network infrastructure.

However, artificial intelligence has changed the story. While organizations are racing to roll out AI tools at scale, the data center industry is discovering that not all workloads have the same infrastructure requirements.

For instance, training a large language model, running real-time inference, supporting enterprise applications and processing business-critical transactions each place very different demands on the underlying infrastructure.

Today, one data center doesn’t need to serve every purpose equally and we’re increasingly seeing that facilities can be both flexible and tailored to specific workload requirements.

The end of the traditional model

Historically, 99.999% uptime was non-negotiable. Data centers have traditionally powered systems like banks, emergency networks and customer-facing digital services, requiring continuous availability.

In these types of environments, where outages could have an extreme impact (from high financial losses to putting real lives at risk) this approach makes sense. Since operators couldn’t always predict which applications would be truly mission-critical, many facilities were built to the highest resilience standards by default.

But AI has changed this. “One-size-fits-all” redundancy isn’t necessary anymore. Different models, training and inference processes each require totally different service levels. For example, facilities for AI training workloads are being designed without backup generators, complex redundancy systems or high-tier architecture.

The good news is that there is a growing understanding of the distinction between environments needed for AI training and AI inference. Training facilities are increasingly being located wherever power is available.

The primary constraints are energy supply, cooling capacity and speed of deployment. In many cases, maximizing compute density and accelerating delivery timelines are more important than achieving the highest possible redundancy levels.

Inference infrastructure presents a different set of priorities. These workloads are often deployed closer to users and support services that people interact with daily. In these scenarios, latency, availability and customer experience become notably more important, creating a stronger case for resilient infrastructure and geographically distributed architectures.

Precision resilience to support an industry under pressure

It’s clear, therefore, that reliability still matters. However, infrastructure requirements vary significantly depending on the service being supported. In today’s age of AI, ‘precision resilience’ should be the focus, e.g., redundancy matching how workloads actually behave, rather than relying on legacy design assumptions.

The key challenge here for operators is determining where resilience delivers genuine business value and where it simply adds cost and complexity.

In a time when developers are facing a huge amount of pressure amid labor shortages, with demand outpacing supply, defaulting to ultra-resilient, high-tier designs for every AI deployment only intensifies challenges.

The industry is also expected to deliver capacity faster than ever before, while battling an ongoing power gap, meaning large-scale developments are increasingly difficult to execute. In this landscape, overengineering infrastructure can have unintended consequences.

Every additional layer of redundancy consumes capital and increases complexity. This is triggering an increased focus on efficiency, not just in terms of energy consumption, but in how capital is allocated throughout a project. Operators are looking to design infrastructure that maximizes the value generated by every watt of available power.

The role of upgradability

As operators move away from this one-size-fits-all redundancy to optimize their bottom line, it’s crucial that their facilities can adapt as workload requirements change.

While inference is expected to account for a growing share of AI demand, the landscape continues to evolve and it’s difficult to predict which workloads, densities and cooling requirements will dominate in the future. Infrastructure that can accommodate changes in compute technologies will be better positioned to support the next generation of AI applications.

Flexibility and fungibility are therefore the new non-negotiables in data center design. How is this made possible? Increasingly, developers are using ‘building blocks’ constructed off-site in factory environments, and then later assembling them on site to create an adaptable facility that can forever evolve, grow and shift.

This approach reduces the need to make every resilience decision upfront and builds with tomorrow’s changes in mind. In the coming years, we will see a shift towards multiple types of facilities, each developed for a different purpose.

These will range from energy-optimized training campuses built close to power sources, to distributed inference sites where uptime and latency directly affect user experience, alongside hybrid environments supporting both AI and traditional workloads. Yet they should all be built with flexibility front of mind to ensure they can evolve as requirements change.

We've featured the best web hosting service.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

• The number of companies using AI is up, but the number of tools per company hasn't grown much
• ONS data also confirms that AI doesn't seem to be having a major impact on headcount
• SMBs could be lacking a clear business use case for AI

New ONS data has claimed that while AI adoption has expanded rapidly across UK businesses, the actual depth of adoption is still limited, with most firms using just a small number of tools to improve existing processes rather than totally overhauling their businesses.

Among the businesses surveyed that had adopted AI, the number of tools they used only increased marginally from 1.4 to 1.6 in the three years leading up to 2026.

The ONS also revealed that, while 60% of larger businesses have used AI to improve business operations, no negative impacts on headcount have been observed.

AI adoption might be growing, but that's not the full story

The data mostly implies experimentation rather than widespread transformation, with just 17% of SMBs (0-9 employees) reporting extensive AI use, however the report warns that many small businesses are merely trying general-purpose AI tools rather than making substantial investments into tech that can truly unlock new levels of productivity.

What's less clear is why SMBs are falling behind their enterprise counterparts, because while 7-14% of all business sizes cited cost as a barrier, 41% of SMBs with 10+ employees said they didn't have any major barrier concerns.

Rather than being held back by costs, infrastructure or capability, the ONS data actually implies that SMBs might actually be lacking compelling business cases to use AI altogether.

All in all, the most recent data confirms that while SMBs haven't exactly ignored AI, they're still stuck in the experimentation stage. The readiness to use automation tech is very much there, but ability to identify a genuine use case is potentially holding back deeper deployment.

AI is now part of the operating system of SaaS. It shapes how products are built, how teams collaborate, and how quickly ideas move from concept to launch.

But speed isn’t the most important shift. The real change is that AI is reducing the coordination cost inside organizations.

Work that once required multiple layers of approvals, handoffs, and alignment can now move more directly between the people closest to the problem. And as that friction drops, something more fundamental starts to change: how companies are structured.

Projects that once demanded large teams, heavy investment, and long development timelines can now be delivered by smaller groups using AI tools to accelerate execution.

The rise of micro-SaaS businesses is one clear example, with small teams able to build and scale products with a level of speed that would have been difficult to imagine a few years ago.

This isn’t just about building faster. It’s changing what scale actually looks like.

From experimentation to infrastructure

Today, AI is embedded directly into product development, engineering, growth, and support. It’s no longer something teams experiment with on the side.

This has brought about a fundamental change: individual contributors can move faster, make decisions earlier, and deliver more on their own.

That has a direct impact on how teams scale.

And increasingly, the companies with an edge are not the ones with the biggest teams, but the ones that can remove friction and make better decisions faster.

Why scale no longer means more layers

Traditionally, growth came with added complexity. More customers meant more people. More people meant more managers, more processes, and more coordination.

At a certain point, coordination becomes a job in itself.

AI starts to break that pattern and bottleneck, freeing up time for quality decisions. When a product manager can analyze user feedback, draft a roadmap, and collaborate more directly with engineering using AI tools, you reduce the need for multiple handoffs. When a growth team can produce, test, and iterate on campaigns faster, execution accelerates without increasing headcount at the same pace.

It doesn’t remove the need for structure. But it does reduce the need for layers whose main role is coordination.

And that opens the door to a different model of scaling: one that is lighter, more direct, and more focused on making the right decisions, not just executing faster.

The return of the “contribution era”

What we’re starting to see is a shift back toward what could be called a contribution-led model. For a long time, SaaS organizations leaned heavily into management structures. That made sense when scaling meant handling more complexity across teams, regions, and products.

Now, as AI lowers the cost of execution, the balance starts to shift again: the biggest advantage AI creates is not productivity but organizational simplification.

Strong individual contributors who can own a problem and drive it to completion become even more valuable. They don’t need to wait for as much coordination. They can test, build, and iterate independently. And they can do it while staying closely connected to the outcome. Importantly, this isn’t about removing managers. It’s about rebalancing the system.

What builder-led really looks like in practice

A builder-led model doesn’t mean everyone is an engineer, and it doesn’t mean structure disappears.

It means the people closest to the work have more autonomy to move it forward.

You see this already across teams:

• Product teams prototyping faster using AI-assisted tools
• Growth teams running more experiments with shorter feedback cycles
• Support teams handling higher volumes while focusing human attention where it matters most

In each case, AI is not replacing people. It’s increasing their speed and range.

And when that happens consistently, the bottleneck shifts. It’s no longer capacity. It’s clarity and decision quality: knowing what to work on, what to prioritize, and where to invest time.

This is where leadership becomes even more important, not less.

Instead of focusing on overseeing activity or managing layers of communication, leaders have to focus on creating the right conditions for execution.

In practice, it often looks like:

• Fewer approval steps
• More direct communication between teams
• More emphasis on outcomes rather than process

Leaders still set the direction and make the hard decisions. But they rely more on capable contributors to carry things forward.

In many cases, the most effective leaders are those who can still contribute when needed, not just coordinate others.

Hiring for ownership, not just specialization

This shift also changes how companies think about hiring.

Specialists remain essential. But, if smaller teams can deliver more, the focus moves toward people who combine expertise with ownership, and have a strong ability to make good decisions in fast-moving environments. There’s growing value in hiring people who can operate with autonomy, make decisions, and adapt as things change.

In a builder-led environment, the question is less “what is your lane?” and more “how effectively can you solve the problems in front of you?”

That doesn’t mean everyone needs to do everything. It means teams benefit from individuals who can connect dots, move across boundaries, and take responsibility for outcomes.

Building smarter, not just bigger

It’s important to stay grounded in how this shift plays out. AI won’t fix weak strategy or unclear thinking, and layering it onto already complex processes can sometimes create new friction rather than remove it.

At Weglot, we've seen teams ship projects with significantly fewer handoffs than two years ago. Marketing can prototype ideas faster, product teams can validate concepts earlier, and engineers spend less time on repetitive tasks.

Our support team is another good example. Over time, they've built a suite of AI-powered tools including a case summarizer, customer profiler, drafting assistant, internal copilot, knowledge base, and AI chatbots. Together, these tools help agents access context faster, learn from previous cases, and resolve more requests independently.

The biggest change isn't speed itself. It's the reduction in coordination overhead, which is ultimately a more sustainable way of scaling.

Smaller, highly capable teams with clear ownership tend to stay closer to the product and the customer and can adapt more quickly when things change. We’re already seeing that in micro-SaaS businesses, but the same thinking applies more broadly.

AI will continue to evolve, but one direction is becoming clear. The companies that will stand out are not necessarily the ones that grow headcount fastest. They’re the ones that stay focused, reduce friction, and make it easier for their best people to build and deliver impact.

We've featured the best website for hiring niche employees.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

What do you mean Heartstopper Forever hasn't actually ended? Alongside the fact that the new movie is now streaming on Netflix, the streaming service has slyly announced a "companion piece" — Heartstopper: Ending on a Hi.

Obviously, we learned exactly what would happen to the gang once the cameras stopped rolling, and were left happy in the belief that Nick (Kit Connor) and Charlie (Joe Locke) definitely weren't breaking up, despite how much their terrible communication threatened to do just that.

So... what is actually coming out way? And when does Heartstopper: Ending on a Hi arrive on Netflix?

What time can I watch Heartstopper: Ending on a Hi on Netflix?

(Image credit: Netflix)

Heartstopper Forever drops on Netflix on July 24, 2026.

As for the exact time, it should be the standard 12am PT release that we saw for the movie.

For global regions, here's when you need to be prepared:

• US – 12am PT / 3am ET
• Canada – 12am PT / 3am ET
• UK – 8am BST
• India – 12:30pm IST
• Singapore – 3pm SGT
• Australia – 6pm AEST
• New Zealand – 8pm NZDT

What exactly is Heartstopper: Ending on a Hi?

(Image credit: Netflix)

Heartstopper: Ending on a Hi is an official 35-minute behind-the-scenes documentary special.

Netflix describes it as a "love letter to the franchise, featuring unseen archival footage, cast interviews, and fan reactions following the final Heartstopper: Forever film."

Don't take the title as a sign of something more to come, though... it's a reference to the very first time Nick and Charlie met in season 1.

Every enterprise I talk to right now has the same complaint dressed up in different language.

Their AI bills are climbing faster than anyone budgeted. Their model invoices look reasonable when viewed on their own.

But somewhere between boardroom approvals and the monthly cloud statements, money is disappearing in ways that nobody can fully explain.

Here’s the central issue that people struggle to understand: the most expensive part of AI isn't always the model itself. It's the infrastructure, orchestration, retries, idle GPUs, oversized context windows, and inefficient routing decisions that sit between a user's prompt and the final response.

This is something I call the hidden tax on AI adoption, and it's growing faster than most organizations realize.

Three numbers that should change how you think

Recently, at FinOps X in San Diego, a Goldman Sachs projection appeared on the main stage. Current enterprise token consumption globally sits at around six quadrillion tokens. The three-year projection: 120 quadrillion. That is not a rounding error… it is a 20x expansion, and it is arriving faster than the governance frameworks to manage it.

The same conference saw our launch of the Tokenomics Foundation (I’m fortunate to be a governing board member). It’s a vendor-neutral body inside the Linux Foundation dedicated specifically to the economics of AI token consumption.

The FinOps community, practitioners who have spent the better part of a decade building discipline around cloud computing spend, recognized that tokens represent a fundamentally different problem. Not a harder version of cloud cost optimization. A different one.

Here is why. When your organization runs a cloud workload, the cost is relatively legible. You provision compute, it runs, and you receive a bill. The relationship between action and expense is traceable. With AI tokens, that relationship fractures across three layers, and most organizations have visibility into only one.

Production → consumption → value: the three layers most teams ignore

The first layer is production. Before any AI model responds to any prompt, your infrastructure has to manufacture the tokens. GPU clusters, inference nodes, autoscaling policies, Kubernetes configurations: these are your token factories. Their efficiency, or lack of it, determines the base cost of everything that follows. A GPU node at 30% utilization is an expensive factory running at a third of capacity.

The second layer is consumption. This is where counterintuitive economics live, and it is what I spend most of my time thinking about. Two organizations can send identical prompts to different coding agents and arrive at radically different costs depending on how they manage context, caching, retries, routing, and the infrastructure supporting inference.

Prompt length, context window usage, caching strategy, and model routing decisions all compound. The common assumption that routing a task to a cheaper model always saves token cost, turns out to be wrong often enough to matter. A routing decision that invalidates a warm cache can make a "cheaper" model call more expensive than the frontier option it was meant to replace. These are second-order effects. They do not appear in standard dashboards but they appear in your monthly bill.

The third layer is value. This is the one that FinOps teams are comfortable with, and the one that matters least until you have the first two under control. Mapping token spend to business outcomes is a legitimate and important discipline. But you cannot govern at the value layer without instrumentation at the production and consumption layers. You are doing math with incomplete inputs.

Why 85% of your AI spend is probably misallocated

Here is a pattern I see consistently. Organizations treat frontier AI models, the most capable, most expensive models available, as their default infrastructure. Every task goes to the same model. Every prompt is constructed the same way. There is no routing logic, no tiering, no architectural distinction between work that genuinely requires the full capability of a frontier model and work that does not.

Based on my observations across organizations deploying AI at scale, roughly 15% of software development tasks actually require frontier model capabilities. The remaining 85% of routine coding, summarization, classification, and retrieval work can be handled by smaller, faster, and less expensive models, if you have the infrastructure to make those decisions intelligently and automatically.

The unlock is not picking better models manually. Manual model selection does not scale and degrades the developer experience by introducing friction at the moment when a developer needs to move fast. The unlock is building infrastructure that makes routing decisions for you: one that understands the task, routes it to the appropriate model tier, evaluates output quality, and escalates if needed. You specify the outcome you need. The system handles the economics of achieving it.

This is the direction the industry is moving, and the organizations that build this capability first will have a structural cost advantage that compounds over time.

The invoice arrives last. And you realize something has gone horribly wrong

There is a phrase I have started using with customers that captures the core problem: the invoice arrives last.

By the time you see the model provider bill, the cost decisions were made weeks ago in infrastructure configurations, autoscaling policies, and prompt architectures that nobody has reviewed since the initial deployment.

The retry logic runs silently when an upstream service slows down. The GPU nodes were reserved for peak traffic that never came. The agentic workflow, where a single user request fans out into dozens of model calls beneath it, each billed separately, none visible in the tool that generated the original request.

These costs do not live in the model invoice. They live in the infrastructure layer, in the consumption layer, and in the gap between how teams think their AI systems work and how they actually behave in production. You cannot govern what you cannot see. And right now, most teams are looking at one layer of a three-layer problem.

When AI goes from copilot to coworker, the stakes multiply

There is a shift underway that makes all of this more urgent. The AI deployments most enterprises built over the last two years were assistants, tools that accelerated individual work by handling the first draft, the next suggestion, and the boilerplate. A human remained in the loop at every consequential step. The economics were bound by how many people were using the tool and how often.

Autonomous agents change the economic profile entirely. When an AI system can receive a goal, build a plan, execute multi-step work, evaluate its own outputs, and iterate to completion without human intervention at each stage, you are no longer running an assistant. You are running something closer to a coworker, one that operates continuously, scales horizontally, and generates token consumption at rates that individual user interactions never approached.

The transition from copilot to coworker has already happened. And the governance implications are significantly more serious. A copilot with poor token economics costs you some efficiency. An autonomous agent with poor token economics runs that inefficiency at scale, continuously, without generating the natural friction that would cause a human user to pause or change approach. Infrastructure discipline and token optimization need to be in place before autonomous workloads scale rather than retrofitted afterward when the bill arrives.

The mandate for infrastructure teams

The right answer to this problem is not more dashboards. More visibility into a system you cannot control is just a more detailed invoice; it arrives with the same lag, and it changes nothing about the decisions that were already made upstream.

What infrastructure teams actually need is control that operates at the layer where costs are determined, not where they are reported. That means autonomous management of GPU and inference workloads, continuously rightsizing to match actual demand rather than peak assumptions, absorbing the bursty consumption patterns that agentic jobs produce, and moving compute capacity across providers when one environment becomes the bottleneck.

This is especially urgent now, because the transition from copilot to coworker does not give you a grace period to retrofit discipline. Autonomous agents do not pause. They do not get frustrated and choose a different approach. They run the inefficiency you built into them at scale, continuously, until something external stops them.

So the next phase of enterprise AI is defined by who can deploy models efficiently. As AI systems become more autonomous and token consumption accelerates, competitive advantage comes from understanding the full economics of AI, and not just the price of a model call.

Because by the time the invoice arrives, the decisions that shaped it have already been made.

Use the best business cloud storage to manage your data.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

Insta360 is the major player in all-seeing 360-degree cameras and is no stranger to innovation, also incubating the first 360-degree drone in 2025, the Antigravity A1.

In another first, Insta360 recently released a dual-lens compact vlogging camera, the Luna Ultra, which is a Wall-E robot lookalike that shoots 8K video, with neat detachable remote screen and mic.

And if a recent post on Weibo by its founder and CEO Liu Jingkang, aka 'JK', is to be taken at face value, the company wants to lead us into a new era of camera altogether.

The Weibo post by JK was picked up by The New Camera and after translation reads:

“Our product vision is to build a ‘Cameraman’ — an intelligent photography robot capable of automatically composing shots and capturing moments like a professional photographer, allowing users to remain fully immersed in the present.

"Today, AI is rapidly evolving, we believe the Cameraman’s 'brain' is becoming increasingly capable, unlocking new possibilities for creative imaging.

"In the long term, the ideal end state for cameras is to progressively become imperceptible, even to ‘disappear,’ until recording no longer consumes effort and becomes a natural part of life.”

The Luna Ultra already resembles a robot and there's a POV head tracker accessory available that synchronizes the camera with your head movement to film where you are looking, somewhat automating its function.

Insta360 hopes fans further caricaturize the Luna Ultra's robot-like features through a design challenge contest with Bambu Lab, too. But as a photographer and, frankly, a human being, JK's post raises some serious questions for me.

Here I am, filming with the Luna Ultra. In future, could Insta360 might make autonomous cameras like these? (Image credit: Future / Tim Coleman)Isn't photography a creative, human pursuit?

JK's post implies that his company's future camera gear could be autonomous photography 'robots' that can compose, shoot and edit photos and videos on the fly, with little to no human involvement.

It sounds like having your own intelligent cameraperson and editor — a portable 'photography robot' much more sophisticated than the recent Beni camera robot — that follows you and is able to capture the moment as you enjoy it and piece together the best parts into a reel, with the entire workflow reliant on AI.

This imagined future might actually sound better than the current reality, to be honest: one where people walk around camera in hand, filming themselves front and center, rather than being in the moment.

And certainly when you're dealing with 360-degree clips, an intelligent editor which can suggest where the action is at any moment and piece together the edit appropriately for you is a huge time saver.

For Insta360 gear specifically, JK's vision for autonomous, intelligent cameras kind of makes sense. But as a photographer and creative, I find the idea of a 'robot' doing the job for me somewhat life sucking. It's the latter part of JK's post, however, that concerns me more.

It's come to light that modders have been able to disable the light that indicates when Meta Ray-Ban glasses are recording video. Meta says extra measures are being put in place to prevent this (Image credit: Future)Is it really OK to record everything?

JK's post describes an 'end state for cameras' being progressively 'imperceptible' and 'a natural part of life'. In other words, roaming cameras that the public eventually gets used to, recording anything and everything. So what about consent?

Discreet camera tech is already causing privacy concerns, with a growing backlash to wearable cameras especially Meta's Ray-Ban smart glasses, after it was discovered that modders could disable the light which indicates that the glasses are recording, and a fix had to be put in place.

Cameras are already commonplace in public, especially tourist hotspots. But for the most they are controlled by people with a general sense of what is and what isn't ok to photograph or record. And when someone crosses the line, it's at least possible to request they stop filming. But in Insta360's imagined future? Autonomous robots filming everything sounds like a privacy nightmare.

I've asked Insta360 for comment and clarification on the post, specifically what exactly an intelligent photography robot looks like, and how the company will address any related privacy issues, and will report back if I get further information.

Adobe is a popular brand when it comes to all things creative, including AI content generation and editing. With its newly improved Firefly tool, Adobe has taken AI content generation a step further with realistic, real-time outputs.

In addition to being highly accurate, Firefly is also one of the most affordable tools on the market today. But is it the best AI video editor for your needs? Read on to find out. In this article, we've put Firefly to the test, examining its features, pricing, ease of use, and overall value for money.

Adobe Firefly: Plans and pricing

Adobe Firefly is available online by clicking here, as well as part of the Creative Cloud suite of apps.

It offers one of the most generous free video editing software we've seen. Instead of capping your monthly or one-time usage, Adobe allows you to generate a limited number of images and videos each day. This means you can continue using Firefly for a longer period, provided you stay within the daily limits.

What we like about the free plan is that Firefly does not lock its image or video editing features behind paywalls. Even on the free plan, you get access to features such as Generative Fill, Background Removal, AI Markup, video upscaling, text-based editing, as well as audio features such as generating soundtracks, speech, text-to-avatar, and more. Unlike other platforms, Firefly's free plan can actually come in handy for limited individual use.

(Image credit: Adobe Firefly)

Its paid plans are also among the most inexpensive we've seen in the category, with the Standard plan starting at $8.49 per month when billed annually, which comes with 2,000 credits.

With this plan, you get access to all the AI models Firefly has to offer, with hard upper limits on the number of images and videos you can generate with each model. Videos, however, are capped at a maximum duration of five seconds, which might not be enough for users looking to generate long-form content. Besides this, you get access to unlimited Firefly Boards, while the remaining features stay the same as those in the free plan.

Next is the Pro plan at $16.99 per month (4,000 credits), billed annually, where you get unlimited access to several AI models such as Gemini 2.5 Flash, FLUX.1 Kontext, and others. The maximum number of images and videos you can generate is also higher than in the Standard plan.

Then there's the Pro Plus plan, priced at $29.38 per month, where you get up to 10,000 credits and higher limits on image and video generation. In this plan, a wider range of models comes with unlimited access, including Gemini 3.0, ChatGPT Image, Runway Gen-4.5, and Gemini 3 Nano Banana Pro 2K.

Finally, there's the Premium plan at $118.93 per month, offering 50,000 credits and unlimited access to almost all AI models except a few. This plan is ideal for large content creation and creative teams.

Adobe Firefly: Features

Right off the bat, what impressed us most about Firefly is its brainstorming and early-stage concepting feature called Firefly Boards. It is essentially an infinite-canvas mood board and ideation workspace that lets you upload your own images, sketches, or stock images, or generate new ones.

You can select multiple assets and combine elements while brainstorming ideas with your creative team. This is similar to Midjourney- or Discord-style iteration, giving remote creative teams a space to flesh out new ideas.

(Image credit: Adobe Firefly)

Moreover, Firefly offers access to more than 10 popular video, audio, and image models, each with its own limits on the amount of content you can generate. Some notable features are Generative Expand and Generative Fill, which let you expand the canvas of an image or video after it has been generated using Adobe's AI engine. Similarly, there's also Generative Remove, which helps you remove elements from a piece of content.

Firefly also throws in a range of audio features, such as the ability to generate speech, translate videos from one language to another, or add text-to-sound effects to your content. However, we were a bit disappointed with its video editing features, such as trimming, arranging, and refining content, all of which are still in beta. Its AI assistant is also at a very early stage and currently supports only one conversation at a time.

You also have the option to upload an image to provide Firefly with a style or structure reference and create a consistent look across a batch of images. Unlike many other AI tools, Firefly allows you to use your own images to train its AI models and create custom models. This is still an early-stage feature, and we expect it to improve over the next year or so.

Firefly has also expanded its multilingual reach and now supports prompts in more than 100 languages, along with translations in more than 20 languages. There's also an option for bulk actions such as background removal, color grading, and cropping.

Its mobile apps have also improved a lot over the past year and now offer native iOS and Android apps with an experience that's very close to the web interface. And of course, Firefly is part of Adobe's broader AI-powered content generation and editing ecosystem, integrating directly with apps like Photoshop and Illustrator.

Adobe Firefly: Interface and in use

Adobe Firefly is by far one of the most improved image and video editing software we have seen in a long time. One of the biggest criticisms of Firefly was that it did not offer any option to edit the images it generated. However, all that has changed with its revamped user interface.

Not only did it generate a pretty high-quality image for us, but it also provided several editing options. We especially liked its Tune feature, which is currently in beta and allows you to change the look of an image by selecting from a list of preconfigured options.

There is also a prompt option, which lets you keep fine-tuning your generated image with follow-up prompts.

(Image credit: Future)

However, if that is too much work, there are several fine-grained controls available. For instance, there is a Fill feature that lets you fill any blank spaces in the image with an element of your choice. Similarly, there's the Remove feature, which lets you remove any element you do not want in the final result.

What we liked the most is its Select feature, where you can select a particular element in an image or video and then type a prompt to edit only that selected portion. This helps avoid unwanted changes in other parts of the image or video and ensures that the edit is applied only to the section you selected.

(Image credit: Future)

The generation interface itself is pretty simple. You'll see a prompt box at the bottom of your screen, along with a panel on the left-hand side where you can control the generation settings.

For instance, you can select the model you want to use, the resolution, the aspect ratio (16:9 or 9:16), and the duration of the video. There is also an advanced setting where you can enter a random seed value to experiment with the AI engine settings.

What we found particularly eye-catching is the prompt enhancement feature, which lets you improve your prompts before generating an image or video. This helps you create more detailed prompts, which in turn result in more accurate outputs.

Adobe Firefly: How we tested

We tried the free version of Adobe Firefly over several days and were quite impressed with the results. We first tried to create an AI image with the following prompt:

"A tired line cook in a stained white apron, frowning with exhaustion, wiping sweat off his forehead while chopping six carrots on a wooden cutting board. Behind him, a dim, cluttered restaurant kitchen with steam rising from a pot. Photorealistic, dramatic low lighting, shot on 35mm film."

As soon as we hit Generate, Firefly got to work and produced a pretty impressive image that closely matched our prompt.

(Image credit: Future)

However, it's worth noting that Firefly is a tad slower compared to the likes of Fliki or Kapwing. That said, the accuracy of the results is worth the wait.

Once the image was generated, we clicked on it to open the editing panel and played around with the settings. For instance, we selected the burning steel pot behind the cook and prompted Firefly to change it to a wooden pot instead.

Rather than simply making the change, Adobe gave us an Edit Strength slider, allowing us to control how strongly the wooden effect was applied to the pot.

(Image credit: Future)

We then tried the Remove feature by selecting a bunch of carrots and asking Firefly to remove them. Adobe once again impressed us, as it accurately removed the carrots without distorting the chopping board around them.

(Image credit: Future)

Lastly, we used the Upscale feature to improve the quality of the image. This took the longest, more time than it took to generate the image in the first place. However, we were once again very satisfied with the final result.

We also generated an AI video using the following prompt:

"A woman in a red coat walks across a rain-soaked city street at night, neon signs reflecting in the puddles, camera slowly tracking alongside her."

(Image credit: Future)

The results were again highly accurate, the colors were vibrant, and we were impressed by how closely Firefly followed the prompt without significantly diverging from the intended output.

Adobe Firefly: Alternatives

Although Adobe Firefly is a much-improved product, it does have a few shortcomings, which is why you may want to consider some alternatives. For instance, Adobe isn't the best choice if you want to generate AI avatars. Although it offers the feature in a limited capacity, the results are not what you'd expect from a dedicated AI avatar generation tool.

In that case, you can try Kapwing, which offers one of the most extensive collections of AI avatars. You can choose from a wide range of characters, including traditional human-inspired faces and more outlandish options such as aliens, Cupid, demons, cyborgs, Dracula, and more.

Kapwing produces some of the best photorealistic AI avatars we have seen in the industry. It is also one of the few tools that can render videos in 4K quality. That said, it is a bit on the expensive side, with plans starting at $16 per month.

If you need a dedicated AI assistant to help with content generation and editing, you can try Fliki, which offers an AI Copilot that understands natural language inputs and helps you achieve your desired results more effortlessly. It also includes licensed YouTube music and advanced video generation options such as stop-motion videos.

Fliki also offers a wider range of customization options compared to Adobe Firefly, such as choosing the format and template, customizing the type of character, selecting the AI model you want to use, and adding voiceovers, captions, and AI avatars.

Adobe Firefly: Final verdict

Adobe Firefly is one of the best-known names in the content creation industry, and its recent improvements put it right at the top of our list of the best AI video editors. Firefly offers more than 10 AI models for image, video, and audio generation, helping you create photorealistic images, videos, and audio with simple prompts.

We were particularly impressed by Firefly Boards, which helps you brainstorm ideas on an endless online canvas in real time while collaborating with your content team. Firefly has also improved its editing interface and tools and now supports follow-up prompt editing, along with features like Generative Fill and Generative Remove.

That said, it has a few drawbacks, such as the lack of mature video editing tools, which are still in beta, and rather limited AI avatar generation capabilities. That said, if you already use Adobe tools like Illustrator or Photoshop, adding Firefly to your repertoire is a no-brainer.

We've tested a range of video makers and editors, including the best video editing software, the best video editing apps and the best free video editing software.

• Rheya combines emergency battery power with software designed to predict sewage spills
• Power failures at pumping stations can turn routine blackouts into environmental pollution
• Rheya battery will predict the underlying causes of spills before they occur

UK startup Rheya Labs has developed backup batteries designed specifically to keep sewage pumping stations running during power failures across Britain.

The company was named one of 20 finalists in Ofwat’s £7.5 million Water Discovery Challenge, run by the Water Innovation Fund.

Each finalist receives up to £100,000 in seed funding, forming part of a wider £7.5 million pot distributed over 20 months.

A different approach to an old problem

Sewage spills are a longstanding, common issue, often blamed on overloaded systems or heavy rainfall.

England recorded around 290,000 sewage discharges from storm overflows during 2025, lasting roughly 1.9 million hours combined across the network.

Power failures at pumping stations are widely recognised as one trigger for these spills, alongside heavy rainfall and ageing infrastructure

Rheya Labs is now working on a system that pairs instant battery backup with software designed to flag faults before they cause spills.

When power cuts strike, Rheya's batteries instantly take over the pumps, preventing raw sewage from spilling into rivers and streets.

The system also applies machine learning to pump data, aiming to predict the underlying causes of spills before they occur.

Co-founder Ben Taylor said sewage pollution remains an ever-growing crisis across the UK, calling power resilience at pumping stations a hidden but major cause.

Predictive analytics may give Rheya the edge

Elsewhere, Sydney Water has trialled sodium-ion battery packs at its Bondi pumping station, offering a cheaper alternative to lithium-ion systems.

That Australian project, backed by a $10.6 million grant, pairs a 30-kWh sodium-ion pack with rooftop solar generation for backup power.

In New York, Viridi's 150 kWh lithium-ion battery system replaced diesel generators, supplying backup power for between 32 and 90 hours.

Unlike those systems, Rheya's technology reportedly combines instant power backup with predictive analytics, potentially spotting faults before failures actually happen.

“With our rapidly changing climate, it is more vital more than ever that we double down on our efforts to create a water system able to support lives and livelihoods,” said Jo Jolly, Ofwat’s Director of Innovation.

“The array of new ideas from new players is astounding. The challenge is getting these ideas delivering impact on the ground, with the sense of urgency these times need.”

Whether Rheya's approach genuinely outperforms established sodium-ion and lithium-ion systems remains unproven outside controlled trials and early pilot deployments so far.

Independent verification of Rheya's spill-prediction claims has not yet appeared in public trials, leaving the early results largely anecdotal for now.

Up to ten finalists, potentially including Rheya, will share £550,000 each next April, funding further testing before wider rollout across the water network.

The success of this project will ultimately depend on whether predictive software matches the reliability already demonstrated by simpler sodium and lithium alternatives.

Who'd have thought that the first big spinoff to come out of The Big Bang Theory (not including Young Sheldon and Georgie & Mandy's First Marriage, as they're Sheldon-adjacent) would focus on awkward comic store owner Stuart (Kevin Sussman)?

Our new heroic lead, on-off girlfriend Denise (Lauren Lapkus), as well as friends Kripke (John Ross Bowie) and Bert (Brian Posehn), are brought back together when Stuart triggers a multiverse Armageddon by accidentally breaking a device built by Sheldon (Jim Parsons) and Leonard (Johnny Galecki).

Theoretically, the new comedy show could run until the end of time. But first things first: when does Stuart Fails to Save the Universe episode 1 arrive on HBO Max?

What time can I watch Stuart Fails to Save the Universe episode 1 on HBO Max?

For US viewers, Stuart Fails to Save the Universe episode 1 will drop on Thursday, July 23 at 6pm PT/ 9pm ET.

Internationally, you're looking out for these timings:

• US – 6pm PT / 9pm ET
• Canada – 6pm PT / 9pm ET
• UK – Friday, July 24 2am BST
• India – Friday, July 24 at 6:30am IST
• Singapore – Friday, July 24 at 9am SGT
• Australia – Friday, July 24 at 11am AEDT
• New Zealand – Friday, July 24 at 12pm NZDT

When do new episodes of Stuart Fails to Save the Universe come out?

(Image credit: HBO Max)

New episodes of Stuart Fails to Save the Universe will make landfall every Thursday in the US and on Fridays everywhere else. Here are the all-important dates you need to know about:

• Episode 1: July 23
• Episode 2: July 30
• Episode 3: August 6
• Episode 4: August 13
• Episode 5: August 20
• Episode 6: August 27
• Episode 7: September 3
• Episode 8: September 10
• Episode 9: September 17
• Episode 10: September 24

• A startup founded by former teenage drone-racing champions won a US Army contract worth up to $500 million for its FPV attack drones
• This marks the Pentagon's largest small-drone contract ever, running up till 2031 even as it grapples with taking on Chinese-built drones and components with a much smaller manufacturing base
• This contract allows Neros to push for a 250,000 sq ft factory targeting one million drones annually by 2028

Neros Technologies, founded in 2023 by former teenage drone-racing prodigies Soren Monroe-Anderson and Olaf Hichwa, has won a US Army indefinite-delivery contract worth up to $500 million for its Archer FPV attack drones.

The contract runs through June 2031 and is among the largest of its kind. It comes off the back of tens of thousands of deliveries to Ukraine under a multinational drone-coalition contract. Neros maintains a Kyiv office supporting the country's war effort, where the Archer serves as a Chinese-component-free alternative to the cheap drones that dominate the market elsewhere.

It marks one of the Pentagon's biggest commitments yet to the cheap, expendable aircraft that Ukraine's war has turned into the defining weapon of modern ground combat, and, increasingly, of the airspace just above it.

What is Neros' Archer Drone?

The Archer drone is arguably the most successful US-manufactured budget FPV drone to date.

With two years of Ukrainian frontline use, deliveries running at a thousand-plus per month, and cumulative shipments in the tens of thousands, Neros has a record no other US maker comes close to matching on volume alone.

The Archer is an 8-inch quadcopter carrying a 4.5-pound payload beyond 12 miles, built entirely without Chinese components: a rare mix that put Neros on the Pentagon's Blue UAS approved list and earned it Chinese sanctions.

The deal for Neros' Archer flows from the Army's Purpose-Built Attritable Systems (PBAS) program, which selected it last November as one of three primary manufacturers tasked with pushing modular FPV drones down to the platoon level, and it lands amid a service-wide procurement lurch: Army officials say they intend to grow annual drone purchases from around 50,000 to at least one million within roughly two years.

Neros is hoping not only to ride the wave but to capture the bulk of it: the endeavor that started with 30 basement-built drones the founders hand-delivered to Ukraine for testing is now aiming considerably higher, with the US military underwriting its ambitions.

There is a reason for this: Neros is the closest thing the US has to an answer to Chinese dominance in the industry. This year, it occupied Project Millennium, a 250,000-square-foot Los Angeles facility roughly 100 times its previous footprint, and it has publicly committed to reaching 1 million drones a year by 2028.

If the promise holds, Neros would single-handedly increase current US drone capacity roughly tenfold: a 2025 estimate put total American manufacturing capacity at approximately 100,000 units a year.

For now, the Pentagon's Drone Dominance initiative, roughly $1.1 billion aimed at fielding some 300,000 low-cost attack drones by the end of 2027, keeps colliding with the fact that most American-made FPVs cost multiples of the roughly $5,000 per unit the military currently wants to pay.

Turtle Beach Command Series KB7: one-minute review

Want a high-performing gaming keyboard with an integrated touchscreen for streaming and seamless customization? Then the Turtle Beach Command Series KB7 could be exactly what you’re looking for.

I replaced my usual mechanical keyboard with this TKL model for a few days, and instantly found it to be a joy to use. The low-profile Hall effect switches supply precise, responsive, and buttery smooth typing, with a muted sound that’s great for focusing in-game or for general day-to-day productivity. It's less ‘clicky’ and snappy than some may like, but the lowkey sound and feel was ideal for me.

Like a lot of the best gaming keyboards, there’s also 8K polling, alongside 0.125 latency for instant response typing. On top of this, you get important features like Rapid Trigger, adjustable actuation and remappable keys.

The true highlight, though, is the integrated 4.3-inch touchscreen. This has built-in controls for streamers, providing quick access to OBS and StreamLabs, but also has shortcuts for apps like Spotify and Steam among others. I appreciated how you could use the screen to view PC performance, in-game stats, or set web page bookmarks. You can also use the touchscreen as a numpad, and it feels highly responsive in use.

I also really like the design of the KB7. Its slim aluminum chassis, tactical media control knob, customizable RGB lighting, and generally high standard of build help it to feel worth its relatively high asking price.

There’s no battery option if you want to use the keyboard wirelessly, largely due to the power required to keep the touchscreen going, but those focused on optimized performance will want to go for a wired option anyway. And all things considered, I think the Turtle Beach KB7 is a fantastic gaming keyboard, and one I just might have to continue using for all my PC gaming needs.

(Image credit: Future)Turtle Beach Command Series KB7 review: price & availability

• $199.99 / £189.99 / AU$349
• Released in May 2026
• Arrived alongside Turtle Beach’s KB5 keyboard

The Turtle Beach Command Series KB7 was released in May 2026 alongside the Command Series KB5 — a full-sized keyboard with a smaller touchscreen and mechanical switches.

Unsurprisingly, then the KB7 is pricier than its sibling, and has a list price of $199.99 / £189.99 / AU$349. It goes without saying that it’s undoubtedly a pricey keyboard, but for a TKL with this much tech, it’s not too far from the competition. For instance, the Corsair K70 Pro TKL, which comes without a touchscreen, will typically set you back about $179.99 / £169.99 / AU$239.

Turtle Beach Command Series KB7 review: specs

Layout

TKL

Switch

Titan Low-Profile Hall Effect

Programmable keys

Yes

Dimensions

15.7 x 10.1 x 1.2 inches / 399 x 256 x 31mm (with palm rest)

RGB or backlighting

Yes (customizable)

(Image credit: Future)Turtle Beach Command Series KB7 review: design & features

• Great standard of build
• Touchscreen with streaming controls
• Customizable lighting

First things first, I have to say that the build and quality of the KB7 impressed me. It has a slim aluminum chassis, stylish and sleek look, and premium-feeling keycaps. The keyboard also has a modular design, meaning you can add on the Turtle Beach KP7 keypad as well as the included wrist rest.

And I found said wrist rest to be a great addition — it’s comfortable, well-made, and ensured that using the keyboard for extended periods still felt comfortable. The only real downside is that it increases the keyboard’s footprint a fair bit, so you’ll need a decent-sized desk if you’ve also got a PC / laptop hooked up to a couple of monitors like I do. It’s also worth noting that the KB7 has a double kickstand, and this also helped to reduce fatigue after furiously typing for multiple work days in a row.

The keyboard also features customizable per-key RGB lighting with Turtle Beach’s Swarm II software, accessible via the built-in touchscreen. In my view, the lighting looks great, especially in low-lit conditions. I also really like how there’s an option to create dynamic effects and sync your lighting setup across multiple devices.

(Image credit: Future)

I touched on it a moment ago, but the stand-out feature of the KB7 is the built-in touchscreen. The real highlight here is that the touchscreen features streaming controls. You can easily access OBS and StreamLabs with a single button touch. It’s a useful inclusion, but if you’re not a streamer, there are options for you too. For instance, you can use the touchscreen to set shortcuts for apps like Steam and Spotify, or for websites you frequently visit — you know, like TechRadar.

What’s more, the Swarm II software enables you to personalize aspects like Rapid Trigger and actuation settings or remap keys. You can actually alter the actuation of every key individually, and while I don’t expect most users will need that level of granular control, the sheer degree of customization is undoubtedly impressive.

Finally, there are a few useful buttons, for things like skipping and rewinding media and muting your mic. I also love the tactile control wheel for adjusting volume, although there’s also a dedicated volume mixer available on the touchscreen.

(Image credit: Future)Turtle Beach Command Series KB7 review: performance

• Fantastic typing experience
• Although lighter, more muted feel won’t suit all
• Great for gaming or general productivity

I typically use a mechanical Keychron mini keyboard, and so moving over to the KB7 felt like quite the jump.

Why? Well, it comes down to how typing feels on this TKL keyboard. The low-profiled Hall effect keys feel phenomenally smooth and buttery to press down on, and they produce very little noise. The more muted sound and feel was very different from my usual mechanical keyboard — which has more of a clicky, snappy profile. I much preferred the KB7 to my usual keyboard, though, and although the more lowkey sound and feel won’t suit all, I think it’s ideal for focusing while gaming or working.

Speaking of, I found the KB7 to be a great performer, whether I was playing games or just carrying out typical tasks at work. It’s an extremely responsive model too, carrying an 8K polling rate and 0.125ms latency. Those kind of numbers are certainly good enough for competitive gamers, and I found the KB7 offered pin[point precision at all times.

The touchscreen is also well-sized, and using commands feels highly reactive and reliable. The touchscreen can also display PC performance info and show in-game stats (such as ammo and games won, for instance).

One thing that’s definitely worth flagging, though, is that this keyboard is wired only. The 4.3-inch touchscreen requires quite a bit too much power to run off a battery, so you do lose some of the flexibility that some rivals supply. Having said that, you’ll want to be using your keyboard wired for the best performance and lowest latency anyway, and as a gaming keyboard first and foremost, I don’t consider its wired-only setup to be a major issue.

(Image credit: Future)Should I buy the Turtle Beach Command Series KB7?

Attributes

Notes

Rating

Design & features

Slim, modular design with handy touch-screen and lots of customizability.

4.5/5

Performance

Smooth, reactive typing, 8K polling rate, wired only, but understandably so.

4/5

Value

Undoubtedly expensive, but stacks up well against rivals.

4/5

Turtle Beach Command Series KB7 review: also consider

SteelSeries Apex Pro TKL Wireless Gen 3
There’s a reason why the SteelSeries Apex Pro TKL Wireless Gen 3 is one of our favorite gaming keyboards here at TechRadar. It boasts a premium build, supplies excellent performance, and its analog switches function very well. It’s pretty pricey, but it’s still a top pick for us. Read our full SteelSeries Apex Pro TKL Wireless Gen 3 review.

Corsair Galleon 100 SD
If you’re looking for a different touchscreen-equipped keyboard, then the Corsair Galleon 100 SD could be a great pick. With strong build quality, built-in streaming controls, and slick hot-swappable MLX linear switches, the 100 SD is a highly capable gaming keyboard. It is very pricey, however. Read our full Corsair Galleon 100 SD review.

(Image credit: Future)How I tested the Turtle Beach Command Series KB7

• Tested across multiple days
• Used for gaming and general productivity
• Tested by an experienced tech reviewer

I spent days testing the Turtle Beach Command Series KB7. I replaced my typical Keychron keyboard with this one, using it every day at work for general productivity. But I also used the keyboard when dipping into a few games here and there.

On top of this, I made sure to exhaust the keyboard’s various features and settings and compared it directly against rivals.

More generally, I’ve spent years testing gadgets here at TechRadar, where I serve as a Senior Reviews Writer. I’ve tested all sorts of gear, from premium wireless headphones like the Sony WH-1000XM6 to TVs like the Sky Glass Gen 2.

• Read TechRadar’s reviews guarantee
• First reviewed: July 2026