TechRadar News

• Half say AI has significantly changed their roles, a quarter are worried about job loss
• Three in five believe their employer has no clear AI plan – or no plan at all
• Training and transparency are the most demanded changes by workers

A new Visier report has uncovered just how prolific AI exaggeration could be in the workforce, raising questions about exactly how useful the technology actually is to most workers.

As many as one in two (48%) employees admitted exaggerating their usage or expertise to colleagues or managers some or most of the time, with nearly as many (45%) feeling pressured to use AI at work.

This comes against a backdrop of major change – more than half (54%) say AI has significantly changed their roles over the past two years. With that in mine, the report details how workers feel they've been impacted, and how much support they're getting from the workplace.

AI is changing roles, and many are lying about how they use it

One in three (32%) admit their employer has no clear plan, and more than a quarter (27%) aren't even sure their employer has any AI plan at all, and that much is clear in that only one-third believe their leaders understand how much AI training and support they need.

Barely one-fifth (21%) would strongly agree that their employer provides sufficient training and support, making upskilling opportunities (32%) the most sought-after thing workers are calling for – a close first, just ahead of greater transparency (31%).

Workers are also worried about AI's impacts on their jobs, with 24% calling for more career advancement opportunities and 18% calling for clearer communication about how it could impact them. In fact, the most frequently identified concern is job loss – a worry raised by one in for (24%).

"People told us that they actually want to be equipped and informed, not shielded or protected from AI-driven change," Research and Customer Value Principal Dr Andrea Derler explained.

The report urges businesses to prioritize training and transparency in order to build worker trust, with which they'll be able to build the productivity gains they'd hoped.

As organizations race to adopt SaaS applications and AI-powered tools, shadow integration has become a more dangerous and often overlooked entry point for modern breaches. Every new application introduces its own identities, roles, and permissions, while every third-party integration creates another trusted pathway into enterprise systems.

There may already be hundreds of agents currently inside the average enterprise, and these deployments are not slowing down. Employees routinely authorize third-party OAuth applications that gain persistent access to business-critical systems, often with broad permissions that remain long after they're needed. At the same time, AI agents and automated workflows are being embedded into critical SaaS platforms like Salesforce, ServiceNow, Workday, and Slack.

Each new connection expands the attack surface, yet few organizations have a complete understanding of how these trusted relationships interact, and cyber criminals have noticed. In fact, SaaS breaches have increased by 300%. These identities often operate with elevated privileges, which makes them attractive targets if compromised.

The threat landscape is expanding faster than traditional security models can track. Without simple, trusted controls that can instantly govern SaaS applications and AI agents, organizations are left with systems that can access sensitive data, interact with business systems and take actions on users’ behalf.

To mitigate the threats, security leaders must be able to know at any moment, which third-party apps are connected to their environment, what those connections can reach, and whether they're behaving normally, rather than finding out after a security incident.

Where SaaS innovation creates security risk

As SaaS sprawl grows, IT teams are dealing with fragmented signals across hundreds of applications, making it difficult to understand what’s connected and where access actually lies.

This threat landscape is expanding rapidly. With projections of over one billion agents in enterprise environments by 2028, organizations are deploying these tools faster than any manual review process. As agents gain access to business systems and integrations, traditional approaches to managing permissions and access are no longer sufficient.

Low-code, no-code and AI platforms are making it easier for anyone to build and deploy autonomous agents, often without security or oversight built in. These agents can inherit real access to business systems and data, while organizations still lack a consistent way to discover them, trace their privileges and enforce guardrails across the SaaS environment.

Recent security incidents illustrate this evolving threat. For example, Obsidian Security disclosed a one-click remote code execution vulnerability in Flowise (CVE-2026-40933), an open-source platform for building AI agents and LLM workflows with more than 52,000 GitHub stars.

The research found that an attacker could fully compromise a server by persuading an authorized user to import a malicious chatflow. The risk includes a combination of multiple factors: a trusted user importing a shared workflow, permissions that allowed custom MCP configurations, the ability to launch external processes, and access to connected credentials, APIs, databases, and cloud services. Together, they created a path to full environment compromise.

Traditional solutions don't discover these integrations, can't evaluate their permissions, and have no visibility into the cross-app activity that happens between connected systems. But that’s increasingly how attackers are escalating access and moving laterally.

The role of visibility across SaaS ecosystems

To maintain the speed of innovation while keeping a security-first approach, IT teams need visibility across users, applications, OAuth grants and permissions so they can understand risk as it changes, not after an action has already taken place. They need to be able to answer questions such as, “Who still has admin rights across multiple apps?” or “Which inactive accounts still have effective access?”.

This is where approaches such as knowledge graphs can offer a different way to understand SaaS risk. A knowledge graph connects entities such as people, accounts, tokens and resources, showing how they relate to one another and revealing context that might otherwise be missed.

Rather than forcing every SaaS application into a standardized permission model, it captures how users, roles, permissions and integrations actually interact across systems, allowing security teams to understand who has access, what they can reach and how that access was granted.

With that foundation, security becomes adaptive rather than restrictive. High-risk or unusual integrations can be automatically flagged or constrained based on factors such as agent behavior or access patterns, while trusted activity can move forward with minimal friction. This allows IT teams to maintain strong control over the SaaS environment without slowing down legitimate business activity.

OAuth and API integrations make this visibility even more important. They can access large amounts of data and connect directly to critical systems, often without the same oversight applied to human users. Traditional tools such as SIEMs and CASBs can struggle to capture this risk because they were largely designed around human activity and security events.

But SaaS risk increasingly lives in the relationships between users, applications, permissions and integrations. A dormant administrative account may appear harmless until it is connected to an overprivileged OAuth integration or an AI workflow that inherits unnecessary access.

The future of SaaS security won't be defined by collecting more alerts or deploying more point solutions. It will be defined by bringing evidence to show where guardrails are needed, who has access, what they can reach, how that access was granted, and how those relationships change over time. Ultimately, the leaders that can turn this visibility into informed control will be best positioned to embrace SaaS and AI innovation for the business.

We've featured the best endpoint protection 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

ChatGPT has been pushing further into becoming a shopping assistant for a while now. You can ask it for product recommendations and it’ll find specific items, show you images, prices and links to buy them. Sometimes it’ll even add contextual details from what it knows about you, like your color preferences or budget, to help personalize its recommendations even more.

I've been testing it over the past few months and I find it a strange way to shop. OpenAI has tried to make the experience as slick as possible, but I still like visiting a retailer’s website, seeing more photos, browsing the rest of the range and checking reviews and product details before I buy anything.

But its latest shopping feature is much more interesting. Because you can now virtually try on clothes within ChatGPT.

This makes a lot of sense. ChatGPT’s image generation capabilities have become much more realistic since its Images 2.5 update. With more natural lighting, convincing textures and a better understanding of how objects should look in the real world. Apply that to shopping and suddenly the idea of virtually trying on a jacket or dress becomes much more appealing.

To use the feature you’ll need to upload a photo of yourself and then look for something you want to buy, with a Try On button now appearing alongside supported shopping results. Alternatively, you can give ChatGPT an image of an item and ask it to show you what you’d look like wearing it.

Now, this sort of virtual try-on technology itself is nothing new. We’ve seen versions of it for clothes, glasses and makeup for years. But I wanted to know whether ChatGPT's newer and more realistic image generation could make the experience genuinely useful.

I find shopping in person pretty overwhelming sometimes. But online shopping comes with its own problems too. Colors can look different in person, textures aren't always what you expect and it's difficult to know whether something that looks great on a model will work on you.

So I decided to put ChatGPT's virtual fitting room to the test.

Trying it on for size

From L-R. The photo of me I initially uploaded to test the try-on feature. Me wearing the jacket IRL that ChatGPT found. ChatGPT's first try-on result. Then the result after I'd asked it to show me what it would look like slightly open. (Image credit: Future / Becca Caddy / OpenAI)

I started with a full body photo of myself taken while I was out shopping and used that as the photo I uploaded after I clicked the Try On button.

I’d seen examples of ChatGPT recreating the original surroundings in the photo when generating try-on images for other people but that wasn't my experience. Despite trying several different items, it consistently placed me against a fairly plain studio-style background.

But I actually preferred that because it meant there was less visual noise and it made it much easier to concentrate on the clothes.

I started by asking ChatGPT to help me find suede jackets. It surfaced several options currently available to buy on the high street and I virtually tried on the ones I liked.

After it surfaced a jacket available in a shop near me, I decided to go and find out in the wild. That meant I could compare ChatGPT’s interpretation with what happened when I actually put the jacket on. And I was surprised by how close it was.

I deliberately picked an oversized jacket because I thought that would make a good test. It’s relatively easy to paste a fitted garment onto somebody. But understanding that something is supposed to hang loosely seemed like a bigger challenge.

I didn't recreate the exact pose or camera angle from the AI-generated image, so this wasn't a perfectly controlled comparison. But ChatGPT did a surprisingly good job of guessing where the jacket sat on my body, its oversized shape and the way the material fell.

It wasn't perfect, but as a way of answering the question “roughly what might this look like on me?”, I can imagine it being helpful. Even just as the first step in the buying process.

(Image credit: Future / Becca Caddy / OpenAI)Changing things up

After trying a few more suede jackets, I wanted to see how well the feature worked when I moved away from the fairly straightforward jeans and jacket combination.

So I asked ChatGPT to find me dresses I could wear to a formal event and started virtually trying those on too.

Again, the results were convincing enough that I could imagine using them to narrow down a long list of potential purchases. One thing I found particularly useful was color. Seeing a garment against my own skin and hair gave me a much better sense of whether I actually liked a shade than seeing it on a retailer's model.

Then I decided to take things one step further. I told ChatGPT to find me something extravagant that I could wear to a 1920s-style ball and tried on one of the more dramatic options.

Even here, the result was remarkably convincing at first glance. The clothing looked like clothing rather than something crudely pasted onto my body, and ChatGPT handled the radically different silhouette surprisingly well.

But by this point I'd also noticed something else sort of awkward was going on. The woman in these pictures wasn't quite shaped like me.

ChatGPT gave me a makeover

The full-body photograph I uploaded was an accurate representation of my proportions. Yet across the generated images, ChatGPT seemed to subtly change them. I looked taller noticeably and more statuesque. My torso appeared slightly shorter and my legs were longer.

You might think that's hardly something to complain about. But that's a huge problem when the entire purpose of the feature is to show me what clothes will actually look like on my body.

We know that generative AI has a tendency to beautify people. I’m convinced this is one of the reasons people make AI images of themselves, because it makes everyone look a little more conventionally attractive. But in a virtual fitting room accuracy matters far more than flattery.

If a dress looks fantastic because the AI has subtly changed your proportions, that's potentially no more useful than looking at the dress on a model in the first place.

(Image credit: OpenAI / Apple / Becca Caddy)Should you use ChatGPT’s try-on feature?

Maybe. I went into this experiment pretty skeptical. I've tried enough AI image tools to know that "realistic" can fall apart very quickly when you look closely, and virtual try-on technology has been around long enough that simply putting digital clothes onto a photograph isn't particularly impressive anymore.

But ChatGPT did do something useful. For me, its biggest strength was color. It might have changed my body, but it kept my skin tone and the color of my hair identical.

Which meant that being able to quickly see several different shades against something approximating my own appearance could genuinely help me decide between them. I was also impressed by its ability to communicate the overall silhouette and feel of an item, particularly with the oversized jacket I was able to compare with the real thing.

So sure, it could take some of the stress out of shopping. Rather than ordering five jackets and hoping one works, I can imagine using virtual try-on to narrow those five down to two.

But I'd still be cautious about treating the resulting images as anything like an accurate prediction of exactly how something will fit. Fabric weight, sizing, movement and all the strange little differences between our bodies are difficult to judge from a generated image. And, based on my experiment, ChatGPT may give you slightly more conventionally flattering proportions in the process.

The robots are among us. They’re carrying passengers through the streets of San Francisco, delivering parcels over the air, and working their way across farms, factories, mines, and industrial sites.

The collision we’re now witnessing between artificial intelligence and the physical world is often described as “Physical AI” – machines capable of perceiving their surroundings, making decisions, and taking action among us in the real world. But every new machine that clocks in has a second job we rarely see. It creates data. Lots of it.

In the US, the Federal Aviation Administration expects the commercial drone fleet to grow from approximately 966,000 aircraft at the end of 2024 to around 1.18 million by 2029. And in factories around the world, there was an estimated 4.66 million industrial robots working in 2024, while another half a million were installed during the year – more than twice the number installed a decade earlier.

Today’s autonomous machines are vastly more capable, but moving all that accumulated experience off them can still be surprisingly manual. In 2026, the fastest way to move masses of data off a machine is still a person carrying a hard drive. And as autonomous machines spread further into our cities and industries, keeping that information moving could become one of the defining infrastructure challenges of the robotics age.

Data runs both ways

Imagine what an autonomous vehicle experiences. Cameras watch the roads from multiple angles, sensors and AI measure and make sense of the surrounding environment, while on-board systems track position and performance of the vehicle itself. Decisions are made and recorded in real-time as pedestrians cross roads, traffic lights change, and thousands of other vehicles and connected technologies move and exchange data around it.

Delivery drones, agricultural machinery, warehouse robots, and autonomous industrial equipment may operate in different environments, but the logic tracks. Every journey or task leaves behind a digital trail of sensor readings, video, telemetry, diagnostics, and operational logs.

And the information doesn’t travel in only one direction. These machines may also need fresh maps, software updates, configurations, AI models, and instructions from the computing infrastructure behind them. What they experience in the physical world flows back toward data centers, while the intelligence needed for their next task travels outward.

Multiply that exchange across millions of machines and an interesting new layer of internet traffic begins to emerge that we’re only just beginning to think about.

The direction of travel was visible as early as 2019, when a landmark university paper on 6G identified mobile robots, drones, and autonomous vehicles would drive demand for ubiquitous wireless intelligence – its vision for the 6G era included networks capable of supporting a 10,000-fold increase in traffic. The US commercial drone fleet alone is now heading toward seven figures, while industrial robots already number in the millions globally.

Perhaps most notable to casual observers, autonomous vehicles are becoming far more prevalent on the streets of our capitals and major cities. In June of this year, machines surpassed 57% of web traffic globally – the first time that bots have outnumbered humans – and they use networks in a very different way. Cloudflare predicts that non-human traffic could eventually surpass human traffic by a factor 1,000 within the next five years.

And a lesson learned by one machine needn’t remain with that machine. There’s an opportunity here that only connectivity can unlock. Google DeepMind’s Gemini Robotics 1.5 showed skills transferring “zero-shot” between physically different robots – a task taught using a two-armed Franka robot was executed by a humanoid, without any data collected on the target machine.

Sit with that implication. A person who masters a technique carries that knowledge around in their head, and when they retire, much of that experience leaves with them. A robot can potentially hand what it has learned to every other robot on the network, which means experience stops being “local” and starts becoming a shared asset across the network, compounding intelligence for the company’s moat.

The physical footprint of robots is spreading outward, but the data traffic they create remains concentrated at central locations. AI may increasingly have wheels, wings, and arms, but their intelligence and interactivity with the world remain still wholly dependent on their relationship with the nearest data center hub, and that’s a problem we now have to solve.

What’s the hold up?

When a robot arrives back at the depot and “clocks out” for a while, it presents an ideal opportunity to recharge, undergo inspection, run self-diagnostics, download updates, and get the data it needs for its next shift. This is when potentially enormous volumes of data, depending on the fleet size, need to be moved both to and from the device.

Sites that house, charge, or service machinery can therefore find themselves functioning as miniature data hubs, dealing with bursts of traffic as machines arrive and depart. And as fleets grow, so does the strain on the network connecting all these hubs. When uploading and downloading take longer than charging or maintenance, the network becomes the operational bottleneck to scale.

A new layer of connectivity

A wired network simply can’t evolve at the same speed as the mobile economy. A fleet operator might need to open a new depot in another city, place drone nests closer to customers, or connect autonomous equipment at temporary industrial sites. Extending fiber to each location would involve trenching, permitting which are often cost and time intensive.

Wireless connectivity offers a faster and more flexible approach. Links can be installed without digging or waiting for permits, allowing operators to connect new sites quickly. Links can also be redeployed or expanded as fleets grow as their physical footprint expands dynamically.

Wireless optical communication (WOC) could provide that high-capacity bridge. Using narrow beams of light, it carries fiber-like capacity through the air between two points with a clear line of sight, up to 20 Gbps across distances of up to 20 kilometers. It can be installed on existing radio masts or buildings in a matter of hours.

For autonomous fleets, that creates a way to rapidly offload telemetry and operational data, upload new maps and models, and return high-value assets to work instead of leaving them waiting at the depot.

WOC isn’t magic though. Optical links still need line of sight, and heavy fog and rain can degrade their connectivity. That’s why serious deployments pair them with an RF path and engineer them for availability rather than simply assuming it. It won’t replace fiber, but it will extend fiber’s reach to the places fiber can’t go, and it’s getting smaller, faster, and more affordable as the technology moves from mechanical to solid-state beam steering with silicon photonics. Light is the answer.

The arrival of electric vehicles taught us to look beyond the machine and consider the infrastructure behind it. Autonomous machines will require us to do the same again, but this time what needs to be replenished isn’t just the machine’s battery – it’s the machine's experience, and from it, the shared intelligence of the entire fleet.

Robots are becoming extraordinarily good at navigating the physical world. We shouldn’t leave the final mile of their learning journey to someone carrying a hard drive across a parking lot.

We've featured the best AI chatbot for business.

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

• AI named as one of the three biggest factors to boost productivity
• Visibility over costs and ROI falls short, study finds
• Legacy tech and skills gaps are ongoing issues

Over half (50%) UK CEOs have credited AI with being one of the three biggest contributors to their productivity improvements over the past year, ranking ahead of more traditional aids like redesigning and simplifying business processes (40%) and reselling workers (38%).

But despite naming AI as one of the biggest influencers on productivity, business leaders aren't exactly sure how it's helping, and only 14% of UK CEOs have full real-time visibility over their AI costs, a new EY report has claimed.

As such, one in three (32%) say they're unable to properly asses AI's ROI, making it difficult for them to actually know whether its impacts are translating into financial value.

AI is probably boosting productivity, but we don't really know how

"UK CEOs now want clear evidence that their investment is delivering stronger productivity and measurable business value," UK&I Managing Partner Silvia Rindone explained, implying that companies need greater visibility across their AI spend and impacts.

However, it's not just difficulties quantifying ROI that's helping non-adopters back when it comes to AI. Legacy systems (34%) and organizational complexity (29%) are still considered major blockers four years after generative AI was considered to have gone ultra-mainstream with the public launch of ChatGPT.

Skills shortages also continue to plague businesses, with three in four (74%) UK CEOs anticipating these gaps to worsen over the next three years – more so than access to capital, proving they can't even buy their way out of difficulties.

While organizations are seeking solid evidence that their investments are delivering measurable ROI in terms of productivity and revenue growth, EY's report emphasizes the fact that they can't overlook the same blockers that have been reappearing in studies over the years – legacy tech and skills shortages.

"Realising value still depends on rigorous analysis, clear strategic choices and disciplined execution," UK&I AI & Data Leader Matthew Ringelheim summarized.

• Control Resonant composer Petri Alanko says every Remedy game has "absolute evil lurking in there"
• Alanko says he and Remedy rely on players' intelligence to understand the "psychological aspect to it all"
• He adds that "Remedy is not a dictating entity; we don’t do walking simulators, and we don’t hold hands"

According to Control Resonant composer Petri Alanko, every Remedy Entertainment title has a sinister aspect at its core, and he relies on the player's intelligence to understand it through his music.

When asked if there is a particular approach to the whole Remedy-verse score and musical landscape in an interview with TechRadar Gaming, Alanko explained that "There’s always the absolute evil lurking in there," with an added psychological aspect.

He said, "It affects instruments, instrumentation," as well as "the overtone series" (a sequence of musical pitches or frequencies that naturally sound above a main base note when it is played), as something so evil it "has a gravity" and "that affects the sounds, too, by bending and separating the overtones from each other."

"Its nature is very prominent, and there is a psychological aspect to it all: when you hear something familiar to be slightly off, it triggers your lizard brain into action, and with some preparation, we’re able to have a very suitable fight-or-flight reaction to introduce some action as a spice," Alanko said. "It is a very, very interesting process to affect people, to sort of kick them into action without actually telling them what to do."

As for how he and Remedy portray that "evil" across the games, he explained that they rely on players' understanding of subtext and context clues to find the meaning of what they're experiencing.

"Remedy is not a dictating entity; we don’t do walking simulators, and we don’t hold hands; we believe in intelligence and the ability to think and deduce something from the facts given," he said. "Same applies to music: a set of hints and a basic feeling may be introduced, but it ends there.

"A good mentalist always lets the audience do the magic. And everyone has seen those really awful marketplace magicians; nobody wants that for a birthday. Remedy is not that. We believe in a thinking, investigating, curious human being."

This applies across all of Remedy's games, whether it's Control, Alan Wake, or Quantum Break; however, Alanko and the studio know they can't regurgitate the same thing over and over again, but players will find a commonality between each series.

"Every new IP adds to the recipe, more or less, and though we don’t exactly want to repeat ourselves, there must be something in common and something familiar to the scores and the soundtracks, and that’s something you can’t really create overnight," Alanko said.

"The same applies with everything: it gets better with time. For instance, we introduced orchestra to Remedy’s sonic identity with Alan Wake, and it was there as is, but also as a slightly mangled version of itself, too. Then there was a very softly played grand piano, with enough compression to turn Manhattan into a peanut, with a reverb that ate half of Helsinki, basically. Quantum Break allowed for much more industrial and electronic flavours to enter the soup, and Control brought musique concrète in."

He added that those three qualities created "What I love to call 'A Remedy Sonic Bermuda Triangle', some top-notch processing," noting that a "huge amount of raw material is originated from acoustic sources, as "There’s very little 'just synths' stuff, although I do love my arsenal."

In the same interview, Alanko has suggested studios should value music more if they care about their games and avoid AI at all costs.

For the past few years, much of the conversation around AI has centered on capability. We have watched it move from generating content and summarizing information to analyzing data, writing code and completing increasingly complex tasks. Now, with the rapid development of AI agents, we are beginning to hand over not just tasks, but portions of workflows and decisions as well.

That progression has understandably raised questions about what happens to people as AI becomes more autonomous, but we are looking at that question the wrong way. The bigger change will be in what we need people to do. When AI tools can handle more of the execution, people have to spend more time thinking about the system around it. We need to ask ourselves important questions to do this:

• What are we actually trying to accomplish?
• What information does the technology need to make a good decision?
• How should work move from one step to another?
• When can AI make a decision on its own, and when does someone need to get involved?

These are questions companies already deal with today, but AI raises the stakes because decisions and actions can happen much faster and at a much greater scale. Take a marketing organization as an example.

We are already at a point where AI can analyze an audience, develop content, adapt that content for different channels, check it against requirements and evaluate the results. It is not difficult to imagine those capabilities becoming connected so that one action automatically triggers the next.

The strategy comes in the coordination

The challenge is that every part of that process can work exactly as designed and still produce a bad result. An audience model could identify the people most likely to respond to an offer while missing the customers the business most wants to develop over time.

A content agent could produce something likely to generate clicks but inconsistent with where the brand wants to go. A system evaluating performance could make a perfectly reasonable recommendation based on the metric it was given, even though that metric does not reflect the larger business objective.

Those are not necessarily failures of the AI. In many cases, the technology has done exactly what it was asked to do. The failure happened earlier, when the objective, context or connections between decisions were not designed well enough. This is where the role of people becomes increasingly important.

Someone has to understand the broader business context and make sure the individual decisions being made across a workflow ultimately support the same goal.

Anyone who has spent time inside a large enterprise has seen a version of this before. Over the past two decades, companies bought specialized technology for almost every business need: CRM platforms, marketing automation systems, analytics tools, project management software, content platforms and countless point solutions. Each solved a real problem, but companies then had to figure out how to make all of those systems work together.

Teams built integrations, moved data between platforms, reconciled conflicting information and created processes to bridge gaps between systems that had never been designed to operate as one. In many organizations, employees effectively became the integration layer, carrying context from one platform or team to the next.

As companies adopt AI across functions, there is a very real possibility that we recreate the same problem, only much faster and with technology that can independently make decisions and take action.

It is relatively easy for a department to adopt an AI tool or build an agent that solves a particular problem. As that happens across an enterprise, however, organizations could quickly find themselves with large numbers of models and agents acting on different data, following different instructions and optimizing for different outcomes.

The individual components may be far more intelligent than the software systems that came before them, but that makes coordinating them even more consequential and increases the importance of having a strategy that extends beyond any individual agent or use case.

Finding the balance is key

Human strategy becomes critical in designing that larger system, but that should not be confused with putting a person behind every AI decision. Requiring someone to approve every output would defeat much of the purpose of creating autonomous systems in the first place.

Companies will have to determine where they are comfortable allowing AI to operate independently, where monitoring is sufficient and where the consequences of a decision justify direct human involvement.

That balance will probably look different across every organization and even across workflows within the same company. A routine decision with clear rules and limited consequences may require very little intervention, while a decision affecting a customer relationship, a brand, significant spending or strategic direction deserves a different level of scrutiny.

There will also be situations where the system simply needs to recognize that something unusual has happened and bring a person back into the process. Determining those boundaries and designing the workflow around them may ultimately matter more than the individual model or agent a company chooses.

Adapt and adjust, as technology changes

The technology underneath all of this will continue to change quickly. Models that seem dramatically better than their competitors today may be overtaken within months, and capabilities that currently feel differentiated will become widely available. Companies cannot assume that access to a particular model will provide a lasting advantage, especially as the gap between competing technologies continues to narrow.

What will be much harder to replicate is the thinking that surrounds the technology: an organization's understanding of its customers and operations, the way its workflows have been designed, the context it gives its AI systems and the judgment it applies to the decisions that matter most. Those things are specific to the business, and they become more valuable as the underlying technology becomes more widely accessible.

Greater AI autonomy therefore does not lead to a future with less need for human intelligence. It creates a greater need for people who can look across the entire system and understand what the technology is being asked to accomplish, whether the pieces are working together and whether the outcome actually makes sense for the business.

We are going to get very good at building AI that can do things on our behalf. The harder challenge will be making sure all of those things add up to something worth doing.

We've featured the best AI chatbot for business.

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

As an experienced deals hunter, I was fully under the impression that, following the recent Amazon Prime Big Deal Days event, we wouldn’t be seeing any particularly noteworthy discounts on coffee machines until Black Friday.

Having recently checked Amazon’s listing, however, it seems I’m very much in the wrong. I like to think that doesn’t happen often, of course, and on this occasion I’m more than happy to be told otherwise.

Yes, I’ve found some superb discounts on some of my favourite espresso machines that either match their Prime Day price, or actually beat it. The Nespresso Vertuo Creatista at AU$697 is a couple of dollars cheaper, while the Sunbeam Origins Sense at AU$475.15 is over AU$80 less than what I saw during Amazon’s recent sale.

It’s also pleasing to see the Philips LatteGo 4400 maintaining its discounted AU$799 price tag. It’s not the best I’ve seen, as it was eligible for an extra AU$175 coupon saving during Prime Day 2.0, but this AU$400 saving is still worth shouting about.

There are a few other coffee machines with big discounts right now, which you can view below. And as ever, make sure to view the full selection on Amazon for something that might be more to your liking. And while I’m on the subject of coffee machines, I have to remind you that Breville currently has a birthday sale live on its website, and TechRadar readers can save an extra 10% on already discounted models thanks to an exclusive coupon code.

Vertuo Creatista capsule coffee machine: was $1099 now $697
We rate the Vertuo Creatista as the best Nespresso machine out there thanks to its use of the Vertuo pods, which come in a range of cup and mug sizes and an integrated milk frother. It performs exceptionally well, too, and will brew your morning cuppa in no time at all.View Deal

Origins Sense Espresso Machine: was $799 now $475.15
This attractive bean-to-cup espresso machine from Aussie brand Sunbeam brews a fine espresso. For me, however, its headline feature is a thermometer built into the steam wand and four temp settings for various dairy milk alternatives, so you can get perfectly texturised milk with ease. It also sports a built-in conical burr grinder for precision grinding, and it has easy, one-button cleaning.View Deal

Origins Dual Espresso Machine: was $999 now $575.45
I’ve not used this exact Sunbeam model, but if it’s anything like the Sense above, you’re in for a treat. The main feature of this model is a dual-boiler system, meaning you can pour an espresso and steam milk at the same time, saving you valuable time in the morning.View Deal

Rivelia (EXAM44057W): was $1599 now $1099
Now available for AU$180 less than Prime Day, this is a great price for an excellent, compact bean-to-cup coffee machine. It scored a full five stars in our De’Longhi Rivelia review, due to it being able to consistently serve up a delicious coffee. There's plenty of customisation on offer, so you can tweak the various preset recipes and it uses the brand's LatteCrema technology for expertly frothed milk — it includes carafes to create hot and cold milk foam.

Also available in black for the same price.View Deal

Vertuo Pop+ w/ Aeroccino. 3: was $349 now $189
The Nespresso machine is a slightly larger version of the highly popular Vertuo Pop. The main difference is a larger 1.1L water tank (compared to 600ml with the Pop) and the water tank can be moved to help you better accommodate the machine in your kitchen. Elsewhere, it uses the extensive Vertuo capsule range, with five different coffee sizes.View Deal

4400 Series Fully Automatic Espresso Machine: was $1199 now $799
As mentioned earlier, this machine did go lower in price during Prime Day, but it’s pleasing to see it’s maintained a sub-AU$800 price tag in the week since. I consider it to be an excellent bean-to-cup espresso machine, as it can make 12 coffee-based drinks, all of which are served perfectly time after time. Plus, it's a great option for smaller households due to it having 2 customisable profiles. However, I have to admit that I wasn't too fond of its milk foaming skills in my LatteGo 4400 Series review.View Deal

Looking for a different day?

A new NYT Connections puzzle appears at midnight each day for your time zone – which means that some people are always playing 'today's game' while others are playing 'yesterday's'. If you're looking for Sunday's puzzle instead then click here: NYT Connections hints and answers for Sunday, October 11 (game #1218).

Good morning! Let's play Connections, the NYT's clever word game that challenges you to group answers in various categories. It can be tough, so read on if you need Connections hints.

What should you do once you've finished? Why, play some more word games of course. I've also got daily Strands hints and answers and Quordle hints and answers articles if you need help for those too, while Marc's Wordle today page covers the original viral word game.

SPOILER WARNING: Information about NYT Connections today is below, so don't read on if you don't want to know the answers.

NYT Connections today (game #1219) - today's words

(Image credit: New York Times)

Today's NYT Connections words are…

• CALAMARI
• MANDARIN
• STATE
• ABBREVIATION
• NICKNAME
• HANDLE
• SHAPE
• WEATHER
• CONDITION
• CONTRACTION
• SITUATION
• BEAR
• ALKYLAMIDE
• STAND
• ELISION
• MELAMINE

NYT Connections today (game #1219) - hint #1 - group hints

What are some clues for today's NYT Connections groups?

• YELLOW: Where we’re at
• GREEN: Put up or shut up
• BLUE: Brief words
• PURPLE: As seen on letters and packages

Need more clues?

We're firmly in spoiler territory now, but read on if you want to know what the four theme answers are for today's NYT Connections puzzles…

NYT Connections today (game #1219) - hint #2 - group answers

What are the answers for today's NYT Connections groups?

• YELLOW: STATUS
• GREEN: TOLERATE
• BLUE: WORD SHORTENINGS, OFTEN
• PURPLE: WORDS MADE USING U.S. POSTAL ABBREVIATIONS

Right, the answers are below, so DO NOT SCROLL ANY FURTHER IF YOU DON'T WANT TO SEE THEM.

NYT Connections today (game #1219) - the answers

(Image credit: New York Times)

The answers to today's Connections, game #1219, are…

• YELLOW: STATUS: CONDITION, SHAPE, SITUATION, STATE
• GREEN: TOLERATE: BEAR, HANDLE, STAND, WEATHER
• BLUE: WORD SHORTENINGS, OFTEN: ABBREVIATION, CONTRACTION, ELISION, NICKNAME
• PURPLE: WORDS MADE USING U.S. POSTAL ABBREVIATIONS: ALKYLAMIDE, CALAMARI, MANDARIN, MELAMINE

• My rating: Hard
• My score: Perfect

Living and growing up outside of the US put me at a disadvantage for today’s purple group — although I am still not sure I would have seen it even if I was based in the States. Kudos if you did.

I managed to get through without making any mistakes, but this was one of those games where I got by on hunches and luck and never had a strong hold on what the group categories might be.

ELISION was a word that I had not heard of before, but ABBREVIATION, CONTRACTION and NICKNAME all seemed to go together, so I took a chance. I have since discovered that it means “omission of one or more sounds, vowels, consonants, or whole syllables from a word or phrase to make speech faster and easier". Yes folks, it’s a good ‘un.

Yesterday's NYT Connections answers (Sunday, October 11, 2026, game #1218)

• YELLOW: PUT TO USE: APPLY, HARNESS, UTILIZE, WIELD
• GREEN: MONTHLY BILLS: CABLE, ELECTRICITY, INTERNET, PHONE
• BLUE: THINGS YOU MIGHT FIND IN A GLOVE COMPARTMENT: GLOVES, ICE SCRAPER, MAP, REGISTRATION
• PURPLE: ENDING IN GAMBITS: EMPLOY, PATRICK, PERUSE, SYNTACTIC

What is NYT Connections?

NYT Connections is one of several increasingly popular word games made by the New York Times. It challenges you to find groups of four items that share something in common, and each group has a different difficulty level: green is easy, yellow a little harder, blue often quite tough and purple usually very difficult.

On the plus side, you don't technically need to solve the final one, as you'll be able to answer that one by a process of elimination. What's more, you can make up to four mistakes, which gives you a little bit of breathing room.

It's a little more involved than something like Wordle, however, and there are plenty of opportunities for the game to trip you up with tricks. For instance, watch out for homophones and other word games that could disguise the answers.

It's playable for free via the NYT Games site on desktop or mobile.

Looking for a different day?

A new Quordle puzzle appears at midnight each day for your time zone – which means that some people are always playing 'today's game' while others are playing 'yesterday's'. If you're looking for Sunday's puzzle instead then click here: Quordle hints and answers for Sunday, October 11 (game #1721).

Quordle was one of the original Wordle alternatives and is still going strong now more than 1,500 games later. It offers a genuine challenge, though, so read on if you need some Quordle hints today — or scroll down further for the answers.

Enjoy playing word games? You can also check out my NYT Connections today and NYT Strands today pages for hints and answers for those puzzles, while Marc's Wordle today column covers the original viral word game.

SPOILER WARNING: Information about Quordle today is below, so don't read on if you don't want to know the answers.

Quordle today (game #1722) — hint #1 — VowelsHow many different vowels are in Quordle today?

• The number of different vowels in Quordle today is 4*.

* Note that by vowel we mean the five standard vowels (A, E, I, O, U), not Y (which is sometimes counted as a vowel too).

Quordle today (game #1722) — hint #2 — repeated lettersDo any of today's Quordle answers contain repeated letters?

• The number of Quordle answers containing a repeated letter today is 1.

Quordle today (game #1722) — hint #3 — uncommon lettersDo the letters Q, Z, X or J appear in Quordle today?

• No. None of Q, Z, X or J appear among today's Quordle answers.

Quordle today (game #1722) — hint #4 — starting letters (1)Do any of today's Quordle puzzles start with the same letter?

• The number of today's Quordle answers starting with the same letter is 0.

If you just want to know the answers at this stage, simply scroll down. If you're not ready yet then here's one more clue to make things a lot easier:

Quordle today (game #1722) — hint #5 — starting letters (2)What letters do today's Quordle answers start with?

• H

• B

• R

• S

Right, the answers are below, so DO NOT SCROLL ANY FURTHER IF YOU DON'T WANT TO SEE THEM.

Quordle today (game #1722) — the answers

(Image credit: Merriam-Webster)

The answers to today's Quordle, game #1722, are…

• HELLO
• BELOW
• RIGHT
• SLANT

Everything fell into place for me today, with every word being the most expected.

In brief, exactly the sort of game I needed after losing my win stream on Saturday.

Thank you, Quordle gods.

Daily Sequence today (game #1722) — the answers

(Image credit: Merriam-Webster)

The answers to today's Quordle Daily Sequence, game #1722, are…

• OPERA
• TULLE
• FEVER
• DEMUR

Quordle answers: The past 20

• Quordle #1721, Sunday, 11 October: LILAC, INLET, TRIAD, SHAPE
• Quordle #1720, Saturday, 10 October: SWEAT, THIRD, RURAL, CLASP
• Quordle #1719, Friday, 9 October: OLDEN, DUNCE, SWATH, HILLY
• Quordle #1718, Thursday, 8 October: MINTY, ROWER, BROOD, RESET
• Quordle #1717, Wednesday, 7 October: PALER, FROTH, DOZEN, ERASE
• Quordle #1716, Tuesday, 6 October: FRESH, SKULK, BATCH, ENSUE
• Quordle #1715, Monday, 5 October: SIXTH, SPEAK, SEDAN, HEIST
• Quordle #1714, Sunday, 4 October: INFER, RAINY, THING, ARENA
• Quordle #1713, Saturday, 3 October: CHORE, FLYER, STICK, PURSE
• Quordle #1712, Friday, 2 October: IVORY, SADLY, ANGEL, LLAMA
• Quordle #1711, Thursday, 1 October: NAVEL, SHEER, BIRTH, TACKY
• Quordle #1710, Wednesday, 30 September: CHOKE, BATCH, MOSSY, EAGER
• Quordle #1709, Tuesday, 29 September: BILGE, MEDAL, LODGE, PORCH
• Quordle #1708, Monday, 28 September: BRUNT, TORSO, STONE, HEATH
• Quordle #1707, Sunday, 27 September: VALET, CHEST, VIGIL, PUREE
• Quordle #1706, Saturday, 26 September: SNORT, BIBLE, LAPSE, CRANE
• Quordle #1705, Friday, 25 September: SWEAR, FLOOD, GUPPY, GROAN
• Quordle #1704, Thursday, 24 September: TOPIC, METRO, SCALP, FREED
• Quordle #1703, Wednesday, 23 September: SHEET, BLAND, JOIST, GUAVA
• Quordle #1702, Tuesday, 22 September: COMFY, LOBBY, DRESS, PANEL

Looking for a different day?

A new NYT Strands puzzle appears at midnight each day for your time zone – which means that some people are always playing 'today's game' while others are playing 'yesterday's'. If you're looking for Sunday's puzzle instead then click here: NYT Strands hints and answers for Sunday, October 11 (game #952).

Strands is the NYT's latest word game after the likes of Wordle, Spelling Bee and Connections – and it's great fun. It can be difficult, though, so read on for my Strands hints.

Want more word-based fun? Then check out my NYT Connections today and Quordle today pages for hints and answers for those games, and Marc's Wordle today page for the original viral word game.

SPOILER WARNING: Information about NYT Strands today is below, so don't read on if you don't want to know the answers.

NYT Strands today (game #953) - hint #1 - today's themeWhat is the theme of today's NYT Strands?

• Today's NYT Strands theme is… In a manner of speaking

NYT Strands today (game #953) - hint #2 - clue words

Play any of these words to unlock the in-game hints system.

• SLING
• RANGER
• FOUL
• GONE
• LONG
• RACON

NYT Strands today (game #953) - hint #3 - spangram lettersHow many letters are in today's spangram?

• Spangram has 12 letters

NYT Strands today (game #953) - hint #4 - spangram positionWhat are two sides of the board that today's spangram touches?

• First side: left, 8th row

• Last side: right, 4th row

Right, the answers are below, so DO NOT SCROLL ANY FURTHER IF YOU DON'T WANT TO SEE THEM.

NYT Strands today (game #953) - the answers

(Image credit: New York Times)

The answers to today's Strands, game #953, are…

• PARLANCE
• JARGON
• VERNACULAR
• LINGO
• DIALECT
• SPANGRAM: TURNOFPHRASE

• My rating: Easy
• My score: Perfect

Another easy game where I got the Spangram second (there are no prizes or badges for that, unfortunately). Very annoying, although I did appreciate how TURNOFPHRASE took a turn across the board.

I was expecting to find the word “slang” on the board, but made do with the non-game word “sling”. Instead, we had the many different ways in which languages can be described.

Personally, I think DIALECT is fascinating, particularly how someone in the same country can ask for a cup of coffee in a hundred different ways depending on where they come from

Yesterday's NYT Strands answers (Sunday, October 11, game #952)

• PAGER
• ICEBOX
• INKWELL
• TYPEWRITER
• TELEGRAM
• SPANGRAM: OLDFASHIONED

What is NYT Strands?

Strands is the NYT's not-so-new-any-more word game, following Wordle and Connections. It's now a fully fledged member of the NYT's games stable that has been running for a year and which can be played on the NYT Games site on desktop or mobile.

I've got a full guide to how to play NYT Strands, complete with tips for solving it, so check that out if you're struggling to beat it each day.

• World of Warcraft: Forever won't receive any new races from retail World of Warcraft
• Blizzard says its "main focus" is to "build off the foundational lore of Warcraft," since Forever exists in its own "continuity"
• It confirms that the stakes won't be as high as retail WoW to begin with, as "You aren’t THE hero in WoW: Forever, you are just A hero"

Blizzard Entertainment has confirmed that it doesn't plan to add any modern World of Warcraft races to World of Warcraft: Forever, since the game exists in its own timeline.

In an interview with TechRadar Gaming, where the WoW: Forever team discussed adding the controversial dual-faction Skyborne race and its plans for the game's future, Blizzard said it doesn't plan to reintroduce any races from Azeroth-based expansions, such as Battle for Azeroth, Shadowlands, or Midnight.

"We have no plans to add races from WoW’s expansions at this time," Blizzard said, meaning fans shouldn't expect new races like Vulpera, Goblins, or Haranir.

And Blizzard's reasoning for avoiding the newer additions to retail WoW? It explained that it wants to build on the original Warcraft game because WoW: Forever essentially exists within its own timeline.

"Our main focus for the game is to build off the foundational lore of Warcraft," it said. "This means first and foremost the RTS [real-time strategy ] games, and we’ve hit on this right from the start with our content that picks up after the events of the new Warcraft 3: Forsaken Kingdom campaign.

"One of the liberating things about this game existing in its own continuity is that we can forge our own path forward and tell stories that serve the world of WoW: Forever first and foremost. It wouldn’t be fair to WoW: Forever and the players who want to see where this world is going for us to spend a ton of time trying to further flesh stories that have already been told extensively in the modern game."

Blizzard added that it plans to "keep our stories very grounded," explaining that while the stakes of modern WoW are "so vast" and "have that gravity and weight because of how far our characters have gone in that world," WoW: Forever will start "in a very different place."

"We may be a part of larger narratives and events, but how we interact with those narratives will be more subtle and grounded, just as was the case in original WoW," the team said, emphasising that "You aren’t THE hero in WoW: Forever, you are just A hero."

It continued, teasing, "That isn’t to say we’ll never go to some of the places that came in later WoW expansions, but we are fairly committed at this time to doing the WoW: Forever version of those places, not simply retreading the versions of these places that we’ve already seen."

Elsewhere in the same interview, Blizzard confirmed that it's actively working to combat real-money trading (RMT) in WoW: Forever.

• MSI’s Prestige N16 Flip AI+ becomes yet another RTX Spark laptop brand
• The top model carries 128 GB of RAM and a 2 TB SSD
• Every configuration uses a 20-core CPU and a 6,144-core GPU

MSI has launched the Prestige N16 Flip AI+, adding a 16-inch convertible to Nvidia's rapidly expanding RTX Spark hardware lineup.

Microsoft, ASUS, Dell, HP and Lenovo have also introduced or announced RTX Spark systems, while Acer has shown its own compact desktop design

That competition leaves MSI differentiating its device through a 128 GB memory option, a 20-core processor and a high-resolution Tandem OLED display.

MSI packs flagship RTX Spark hardware into a convertible

The Prestige N16 Flip AI+ combines Nvidia's 20-core RTX Spark N1X processor with a GPU containing 6,144 cores across all three configurations.

Its 16-inch Tandem OLED touchscreen delivers 3,840 × 2,400 resolution, a variable refresh rate between 30 Hz and 120 Hz, and peak HDR brightness of 1,000 nits.

MSI fits the hardware into a convertible chassis weighing 1.9 kg, with dimensions of 357.7 × 254.3 mm and thickness ranging from 13.9 mm to 16.9 mm.

Under the hood, a 99.9 Wh battery supports charging at up to 140 W, although MSI has not specified battery endurance or the time required for a full recharge.

The Prestige N16 Flip AI+ comes in three configurations, with memory and storage capacity accounting for the differences between them.

The base model offers 32 GB of RAM and a 1 TB SSD, while the midrange version increases memory to 64 GB without changing storage capacity, and the top configuration doubles memory again to 128 GB and increases storage to 2 TB.

MSI joins Microsoft, Dell and other RTX Spark laptop makers

Microsoft's Surface Laptop Ultra, ASUS's ProArt P14 and P16, Dell's XPS 16, HP's OmniBook Ultra 16 and Lenovo's Yoga 9n 2-in-1 already give buyers several RTX Spark laptop options.

Their approaches differ, with Microsoft's machine using a 15-inch mini-LED display, while ASUS focuses its RTX Spark models on creator-oriented ProArt laptops.

Dell uses the XPS 16 for its RTX Spark offering, while HP has placed the processor inside its larger OmniBook Ultra 16.

Lenovo takes a convertible approach similar to MSI, with its Yoga 9n combining RTX Spark hardware and a 360-degree design.

Acer has also entered the RTX Spark ecosystem with a compact desktop design, although that product does not compete directly with MSI's convertible.

MSI's main distinction is its combination of a convertible form factor, a 128 GB memory option and a high-resolution Tandem OLED display.

The company has opened preorders through authorized retailers, with availability varying by market.

In the United States, the laptop starts at $3,299, while UK buyers pay £3,099 for the 32 GB model with 1 TB storage.

The British price rises to £4,099 for the 64 GB version, a £1,000 increase over the entry configuration, and Eurozone pricing starts at €3,899.90 for the 32 GB model, rising to €5,099.90 for the 64 GB version, and the highest configuration costs €7,999.90, combining 128 GB of RAM with 2 TB storage and costing more than twice as much as the entry model.

Via Notebookcheck

It's Vision's time to shine. After playing a supporting role in myriad Marvel movies and TV shows, the superpowered synthezoid is finally set to star in his own project in VisionQuest.

Set around a year — in-universe, that is — after WandaVision, the Disney+ TV Original will reunite us with Vision (Paul Bettany), who's suffering from an identity crisis following the events of WandaVision's finale. Searching for meaning, he consults the A.I. programs in his programming for guidance and advice, including Ultron (James Spader), aka the villainous android last seen in Avengers: Age of Ultron.

Soon, though, Vision is forced to go on the run with Tommy Shepherd (Ruaridh Mollica), a mysterious individual who claims to be one of Vision's reincarnated twin sons, when nefarious forces come gunning for the highly-advanced tech that Vision possesses.

With that somewhat lengthy story synopsis out of the way, we can move onto the real reason for you clicking on this article. That being: when will every episode of VisionQuest be released on one of the world's best streaming services?

What are the release dates for every episode of VisionQuest?

Paul Bettany is joined by MCU newcomer Ruaridh Mollica in the final Marvel TV show of 2026 (Image credit: Marvel Television/Disney+)

The penultimate Marvel Phase 6 TV series will arrive with a two-episode premiere on Tuesday, October 13 at 6pm PT / 9pm ET in the US and Canada.

Originally, VisionQuest was supposed to make its worldwide debut on Wednesday, October 14. However, Marvel has since announced that it'll be released early in North and South America. The rest of the world will have to wait until October 14 to see them.

As for when the other six chapters should be released, Marvel has also revealed that episodes 3 and 4, and chapters 5 and 6, will launch together in the two weeks that follow. After that, episodes 7 and 8 will arrive independent of each other.

Confused? Don't be. Check out the list below for each episode's confirmed launch date:

• VisionQuest episode 1 — October 13 / 14
• VisionQuest episode 2 — October 13 / 14
• VisionQuest episode 3 — October 20 / 21
• VisionQuest episode 4 — October 20 / 21
• VisionQuest episode 5 — October 27 / 28
• VisionQuest episode 6 — October 27 / 28
• VisionQuest episode 7 — November 3 / 4
• VisionQuest episode 8 — November 10 / 11

What time can I watch VisionQuest?

We'll see human versions of AI programs F.R.I.D.A.Y, J.A.R.V.I.S, U, AND D.U.M-E in VisionQuest (Image credit: Marvel Television/Disney+)

As I mentioned above, every episode of the Marvel Cinematic Universe (MCU) show will come out at 6pm PT / 9pm ET in North America.

Live elsewhere? Here's when new installments of VisionQuest are expected to premiere in your time zone:

• US — 6pm PT / 9pm ET
• Canada — 6pm PT / 9pm ET
• UK — 2am BST
• India — 6:30am IST
• Singapore — 9am SGT
• Australia — 11am AEST
• New Zealand — 1pm NZST

It's worth noting that the above times will change in the weeks ahead.

Take on our Vision-based Marvel quiz!

Think you're a Marvel super-fan? See if you can get 100% on our tough VisionQuest quiz before the Paul Bettany TV show is released on Disney+!

Indeed, with the clocks going back in the northern hemisphere, as well as forward below the equator, some episodes will be released earlier or later than the aforementioned times. As and when they come into effect, I'll update this section.

For now, check out my VisionQuest review, which covers all eight episode (don't worry, it doesn't contain major spoilers). After that, read up on everything we know so far about VisionQuest and find out how to watch the Marvel movies in order if you're planning a Vision-specific re-watch before VisionQuest's arrival.

The AI boom is quickly reaching the physical limits of grid capacity and thermal dissipation. The electric grid is stretched as ever, causing longer connection waits for AI data centers, and conventional air and liquid cooling systems struggle to meet the demands of rapidly growing facilites.

As they continue growing at a rapid pace, AI data centers now face a mix of challenges, from computing power outpacing GPU interconnects’ data transfer abilities, to volatile load swings stretching grid power factor correction (PFC) systems and rack power density overloading traditional cooling systems.

Let’s break down the major thermal and electrical bottlenecks slowing the AI boom and how companies are innovating their way out.

The thermal wall (Rack-level friction)

Standard server racks generate a lot of heat, and high-performing GPU clusters generate even more, pushing thermal limits further than ever. Modern GPU racks exceed 50 to 100 kW of power consumption per cabinet, making traditional air and liquid cooling methods insufficient.

Standard cooling systems keep clusters operating between 65 and 85 degrees, but 90 to 100 degrees trigger performance throttling to prevent permanent GPU damage. With GPU racks drawing massive electrical energy, which converts into heat, preventing them from hitting throttling limits has become harder than ever.

Traditional air cooling systems weren’t built for rack densities surpassing 30 to 40 kW, which AI clusters easily surpass, with 100 kW or more becoming the norm. Maintaining airflow at high densities to cool 100 kW+ power racks requires running power-hungry fans at extreme speeds. This saps electrical energy and worsens the Power Usage Effectiveness (PUE) for the data centers. To cool mega AI clusters with air alone, operators would need “hurricane-grade” airflow, which is impractical. Then how about liquid cooling?

Liquid cooling relies on fluid mechanics, where cold plates use microchannels to force turbulent flows directly over chips to keep them cool. Keeping enough volume of fluid flowing through these microchannels requires massive pumping power, an increasing constraint for data center operators. To mitigate this challenge, operators are pivoting to direct-to-chip cold plates mounted directly on the processor, rather than over it, to keep it cool while consuming less pumping power.

(Image credit: Microsoft)

Phase-change mechanics are also helping GPU clusters cool more efficiently. Here, the clusters use the latent heat of liquid vaporization or solid-liquid transitions to absorb the massive heat spikes from high-density GPU racks. Specialized dielectric fluid comes in direct contact with the hot GPUs and then boils into vapor to instantly pull away heat. The vapor then rises, reaches a condenser, and converts back into liquid droplets via gravity, then drips back into the pool to repeat the process. This process is known as two-phase immersion cooling, and it reduces cooling energy by up to 40%.

Some data center operators are pivoting to closed-loop liquid cooling systems, where recirculated fluid cools the GPUs with little to no evaporation loss, removing the need for frequent top-ups.

For example, Microsoft, a top-three data center operator by capacity, has implemented a closed-loop system for newer data centers, where water is filled once and circulated to servers to absorb heat, moved to air-chilled coolers to cool down, then recirculated to absorb heat without needing fresh supplies. This system is costly and will take some time to roll out at scale, but it demonstrates how companies are innovating against thermal bottlenecks.

The interconnect and board-level limits

GPU clusters rely on electrical copper or optical interconnects to share data, memory, and workloads. Electrical copper interconnects have long been the standard, but high GPU workloads have pushed them to the limit.

Electrical signals lose strength over long distances (greater than two meters), with usable reach halving every time bandwidth requirements double, and massive GPU clusters have made the communication distances longer than ever. Closely packed copper interconnects also generate electromagnetic interference that disrupts other signals, requiring complex equalization systems that drain more power.

There’s also the issue of power loss across copper plates on GPU circuit boards (due to resistance). The more current transmitted through copper plates, the more heat is generated, requiring more cooling energy that itself is already a bottleneck.

With electrical interconnects hitting their absolute limits, data center owners have little choice but to pivot to optical interconnects, which convert data into light pulses that travel through fiber optic cables with minimal loss over long distances. These cables are way thinner than copper bundles, creating more airflow paths in dense, heat-generating GPU racks.

Companies are even moving from pluggable fiber optic cables to silicon photonics, or co-packaged optics (CPO), where optical circuits are embedded directly on GPUs to transmit light over long distances. For now, optical interconnects are much costlier than electrical copper, costing up to 7x more in some GPU setups, but there’s little choice but to absorb this cost in the short term because the latter has reached physical limits. On the bright side, as more companies adopt optical interconnects and invest in mass production, the costs will likely reduce.

The grid & substation crisis

Lastly, electric grid connection is the primary challenge affecting AI data centers. Modern electric grids weren’t designed to handle the rapidly increasing loads of AI data centers, with 30 GW of new capacity added worldwide from 2021 to 2025 alone. Rapidly growing additions have stretched the supply chain for electric grid components and introduced engineering challenges that grid operators must work around.

For instance, high-voltage transformers that ensure safe voltage delivery to GPU clusters are in short supply. Wait times have spiked from 6 to 12 months as of 2020 to two to four years currently, much longer than expensive mega GPU clusters can wait for their operations to get powered.

Existing power factor correction systems on grid networks have become strained by massive load draws from new AI data centers. GPU clusters shift power demand within seconds, with one second of low inference demand drawing stable power from the grid and the next second drawing excessive power because of usage spikes. These fluctuations create complex harmonics that overheat transformers and affect the grid, needing grid operators to invest heavily in mitigation systems.

AI clusters are scaling to gigawatt levels, pushing voltage step-down switchgear to its most tolerable limits. Larger transformers, needed to serve these gigawatt-level clusters, lower electrical impedance, forcing potentially high short-circuit fault currents down a line. If a short circuit ever occurs, magnetic fields exert tough mechanical forces on switchgear busbars, causing them to bend or rip free from their mechanical support. What comes next is a structural failure, which grid operators or isolated data center power networks can’t let happen.

Companies have turned to various solutions to counter these electrical bottlenecks. For one, there has been a major turn to solid-state transformers, which use high-frequency power electronics and semiconductors to step down voltage, not magnetic fields like conventional transformers.

(Image credit: Docan Power)

High-frequency electronics enable solid-state transformers to actively regulate voltage and filter harmonics, leading to fewer fluctuations. They are 97 to 99% efficient, compared to 95% to 97% for conventional transformers, and that 1 to 2% difference adds up to a lot in gigawatt-level data centers. Better yet, they are much smaller and lighter than conventional transformers, enabling faster logistics and deployment.

However, SSTs introduce their own challenges. The internal high-frequency components and chips powering them are expensive, often five times or more per kVA than low-frequency alternatives. They also require multiple power conversion stages (AC-DC and DC-AC/DC-DC), and tiny losses compound during these conversions, making it difficult to hit peak efficiency. Internal components generate heat and age faster, requiring more active maintenance than conventional transformers.

Cost is also an issue, as SSTs are currently much more expensive to build and maintain than magnet-based transformers. However, with more investment in mass production, costs will likely reduce over time, allowing AI GPU clusters to adopt them faster. Siemens, the world’s second-largest transformer manufacturer, recently partnered with fellow German tech firm Reinhausen to mass-produce a modular solid-state transformer for AI data centers, and that’s just one example.

Because SSTs can’t be immediately adopted at scale, and conventional transformer shortages have contributed to long wait times for electric grid owners, AI data center owners have been forced to take matters into their own hands. Some have built their behind-the-meter power generation and substation systems to power GPU clusters as they wait to get connected to local grids. xAI, the Elon Musk-led AI company, added 59 portable natural gas turbines, with a combined capacity of over 400 MW, to power its Colossus data center in Memphis, Tennessee, while it waits for a grid connection.

Some data center owners are also looking at Small Modular Reactors (SMRs), where compact nuclear fission reactors generate power on-site, but that’s still mostly in the conceptual stage at this point. SMRs face various limitations, particularly very high costs compared to gas turbines or renewables.

Still, many other data centers, both in early and late construction phases or completed, remain in long queues for electrical grid connections. For instance, in Northern Virginia, which hosts the world’s largest cluster of data centers (570 and gradually increasing), new facilities face wait times of up to 14 years for a grid connection.

US-based electrical grid operators have invested massive amounts in upgrades ($208 billion in 2025 alone and $1.1 trillion in planned investments from 2025 to 2030, per Edison Electric) but still struggle to meet exploding data center demand. More investments are needed to address electrical grid bottlenecks, and so are new technologies that can reduce data center energy consumption to help distribution networks keep up.

The reality of next-generation compute deployment

With technical details, we’ve explained the grid, thermal, and GPU interconnect bottlenecks stretching the next-generation AI compute deployment boom to its limits. These are serious physical and technical challenges that require innovative approaches to address, and fortunately, there’s been no shortage of these approaches.

We’ve outlined examples like electric grid operators turning to solid-state transformers that allow active voltage regulation and harmonic filtering to keep power systems stable, or some GPU cluster operators building their own power generation systems while they wait for local grid connections.

For thermal bottlenecks, GPU cluster operators are leveraging direct-to-chip cold plates mounted directly on GPUs, instead of over them, to stay cool while consuming less energy. Phase-change mechanics also help, where liquid vapor absorbs heat from GPU racks, pulls the heat away, and recondenses back into liquid to repeat the process, reducing energy needs by up to 40%.

Microsoft has tried to reinvent the wheel by implementing a closed-loop liquid cooling system, using liquid coolants continuously recirculated in a sealed pipe network with minimal top-ups.

However, closed-loop cooling systems are more difficult to expand than open-loop systems as GPU needs grow, and they can be louder under heavy loads as pump motors and radiator fans run simultaneously. At a time of growing public backlash against new data centers, environmental noise matters too, not just technical limitations. Closed-loop systems are also more expensive, but costs can reduce over time as more companies adopt them.

For interconnect limits at the circuit board level, companies are switching from electrical copper to fiber optic cables because they transmit data faster and more reliably over long distances, even though they cost more. GPU manufacturers are working on embedding optical circuits directly on chips to transmit light even more efficiently than cables.

Despite these challenges, AI data centers continue to grow at a rapid pace, with 66 GW of new capacity currently under construction in North America alone.

Looking to the future, it's clear a lot is happening in this area, and it’s exciting to watch the various approaches companies have adopted to surmount the electrical, thermal, and interconnect bottlenecks.

The Samsung Galaxy Z Flip 8 arrived earlier this year boasting a familiar trick: it folds into a pocket-friendly square and places a surprisingly useful screen on the outside.

With this new model, Samsung has kept the basic Samsung Galaxy Z Flip 7 formula intact, but shaved down the body, improved the hinge, added more processing power, and made the cover screen easier to live with.

The displays, camera hardware, battery, and charging have barely moved, though, leaving last year’s Galaxy Z Flip 7 looking far from obsolete — it's also now much cheaper than its newer sibling.

To help you make sense of what's changed and which model is best, we've weighed up the price, design, displays, cameras, performance, software, and battery life of these two foldable phones to see which deserves your money.

As a disclaimer, we've fully reviewed the Galaxy Z Flip 7, but not the Galaxy Z Flip 8. We have spent time with the former model, though, so we've based our judgements on both hands-on experience and on-paper specs.

Samsung Galaxy Z Flip 8 vs Samsung Galaxy Z Flip 7: specs comparison

Before we dig into the details, here’s an overview of both phones’ key specs:

Samsung Galaxy Z Flip 8

Samsung Galaxy Z Flip 7

Dimensions:

85.7 x 75.4 x 13.1mm folded; 166.9 x 75.4 x 6.1mm unfolded

85.5 x 75.2 x 13.7mm folded; 166.7 x 75.2 x 6.5mm unfolded

Weight:

180g

188g

Display:

6.9-inch Dynamic AMOLED 2X main; 4.1-inch Super AMOLED cover

6.9-inch AMOLED main; 4.1-inch AMOLED cover

Refresh rate:

1–120Hz main; 60/120Hz cover

Up to 120Hz on both displays

Peak brightness:

2,600 nits

2,600 nits

Chipset:

Exynos 2600 / Snapdragon 8 Elite Gen 5 for Galaxy, depending on region

Exynos 2500

RAM:

12GB

12GB

Rear cameras:

50MP main, 12MP ultra-wide

50MP main, 12MP ultra-wide

Front camera:

10MP

10MP

Battery:

4,300mAh

4,300mAh

Charging:

25W wired, 15W wireless

25W wired, 15W wireless

Storage:

256GB / 512GB

256GB / 512GB

Samsung Galaxy Z Flip 8 vs Samsung Galaxy Z Flip 7: price and availability

Samsung announced the Galaxy Z Flip 8 in July 2026, and the phone went on sale on August 7. Prices start at $1,199.99 / £1,149 / AU$1,949 for the 256GB model, rising to $1,399.99 / £1,319 / AU$2,249 for 512GB.

The Galaxy Z Flip 7 arrived in 2025 at $1,099 / £1,049 / AU$1,799 for 256GB, while the 512GB version launched at $1,219 / £1,149 / AU$1,999.

Samsung is therefore asking an extra $100 for the entry-level Z Flip 8, despite retaining the same storage capacity. The gap grows further with the 512GB models, and the older Flip 7 is likely to become even more tempting as retailers clear stock and discounts begin to appear, especially around sales events like Black Friday.

Given the global memory shortage and rising inflation, Samsung's decision to raise the price of the Galaxy Z Flip 8 makes some sense, but it does mean the Z Flip 7 — which is still a great phone — represents even better value in 2026.

Samsung Galaxy Z Flip 8 vs Samsung Galaxy Z Flip 7: designThe Samsung Galaxy Z Flip 8 FutureThe Samsung Galaxy Z Flip 7Future

Samsung has focused on making the Galaxy Z Flip 8 a little easier to carry and use, rather than giving it a big redesign. At 180g, it's 8g lighter than the Flip 7, while its 6.1mm unfolded body is 0.4mm thinner.

Fold the two phones shut, and the difference grows slightly, with the Flip 8 measuring 13.1mm compared with 13.7mm for its predecessor.

The revised Armor FlexHinge is the more useful change. Samsung has reduced friction inside the mechanism and widened the inner edges so that your fingers have more room to open the phone.

In our early hands-on testing of the Galaxy Z Flip 8, the phone felt noticeably easier to open and close than the previous model — a small refinement that should become more valuable when you repeat the motion dozens of times each day.

Otherwise, the family resemblance is strong. Both phones use the same flat-sided clamshell shape, with the large cover screen wrapping around the two rear cameras.

The Flip 7 already felt sturdy and polished, although its familiar Galaxy styling could look rather restrained beside some rival foldables.

Winner: Galaxy Z Flip 8

Samsung Galaxy Z Flip 8 vs Samsung Galaxy Z Flip 7: displayThe Galaxy Z Flip 8's display FutureThe Galaxy Z Flip 7's display Future

Samsung has carried both of the Flip 7’s displays straight over to the Galaxy Z Flip 8. Each phone opens to reveal a 6.9-inch, 1080 x 2520 AMOLED panel with a 1–120Hz adaptive refresh rate, while the outside is covered by a 4.1-inch, 948 x 1048 FlexWindow. Peak brightness remains 2,600 nits.

That leaves little room between them on paper, although this is hardly a weakness for the older phone. In our review of the device, we awarded the Flip 7 a perfect 5/5 for its displays, finding the inner screen bright and clear even outdoors.

The Flip 8 receives some cover-screen improvements, but the main changes concern how Samsung’s software uses the FlexWindow rather than the panel itself. We’ll cover those additions later.

Both phones offer two excellent screens, and there is no convincing hardware reason to choose one over the other.

Winner: Tie

Samsung Galaxy Z Flip 8 vs Samsung Galaxy Z Flip 7: camerasThe Galaxy Z Flip 8's rear cameras FutureThe Galaxy Z Flip 7's rear cameras Philip Berne / Future

Samsung has left the underlying camera hardware unchanged — you might be sensing a theme emerging here.

Both phones combine a 50MP main camera with a 12MP ultra-wide and a 10MP selfie camera, with no dedicated telephoto lens for longer-range shots.

In our testing, the Flip 7 produced strong results with more natural-looking colors than some older Samsung phones and an excellent Portrait mode.

Its folding design also lets you use the cover screen as a viewfinder for sharper selfies from the 50MP rear camera, although the lack of optical zoom limits its versatility.

The Galaxy Z Flip 8 directs its attention towards video.

Camcorder Grip places the recording and zoom controls within easy reach when the phone is held sideways, while Super Steady’s Horizontal Lock can keep footage level as the handset moves or rotates.

Our early hands-on found the latter feature particularly effective, although more aggressive stabilization can crop the frame and reduce image quality.

Winner: Galaxy Z Flip 8

Samsung Galaxy Z Flip 8 vs Samsung Galaxy Z Flip 7: performance and softwareThe Galaxy Z Flip 8's cover screen FutureThe Samsung Galaxy Z Flip 7's cover screenZachariah Kelly / TechRadar

This is where the Galaxy Z Flip 8 begins to put some meaningful distance between itself and its predecessor.

US and Australian models of the Z Flip 8 use Qualcomm's Snapdragon 8 Elite Gen 5 for Galaxy chipset, while the UK version of the phone receives Samsung’s Exynos 2600, replacing the Exynos 2500 inside the Flip 7. The latter chipset features in every version of the Flip 7, in every region.

The older phone still handles everyday apps, split-screen multitasking, and Samsung DeX without much trouble, but its performance was one of the weaker parts of our review. The Flip 8 has considerably more headroom for gaming, demanding apps, and Samsung’s growing collection of AI features — even if your version of the phone uses the Exynos 2600, and not the Snapdragon 8 Elite Gen 5.

Its software is also better suited to the folding design. Android 17 and One UI 9 add a proper FlexWindow app tray and let you rearrange widgets directly from the cover screen.

Both phones receive seven generations of Android upgrades, but the Flip 8 naturally starts one version later and offers a smoother cover-screen experience from day one.

Winner: Galaxy Z Flip 8

Samsung Galaxy Z Flip 8 vs Samsung Galaxy Z Flip 7: batteryThe Samsung Galaxy Z Flip 8 FutureThe Samsung Galaxy Z Flip 7Future

Samsung has kept the same 4,300mAh battery in both phones, alongside 25W wired and 15W wireless charging.

Neither model comes close to the fastest-charging foldables, and the Flip 8 gains no extra capacity to offset its more powerful hardware.

The Flip 7 generally made it through a full day in our review, although heavier photography or gaming could leave it needing an evening top-up.

The Flip 8’s newer chipset should use that capacity more efficiently, but the improvement is unlikely to transform the experience. These remain one-day phones with relatively slow charging, leaving little to separate them.

Winner: Tie

Samsung Galaxy Z Flip 8 vs Samsung Galaxy Z Flip 7: verdict

The Samsung Galaxy Z Flip 8 (Image credit: Future)

The most revealing thing about this comparison is how mature Samsung’s Flip formula has become.

The Galaxy Z Flip 8 improves the parts you touch every day — weight, thickness, hinge action, performance, and cover-screen software — while leaving the displays, camera hardware, battery, and charging largely untouched.

Given these upgrades, the Flip 8 is a better phone for someone buying their first Flip, especially if they plan to keep it for several years.

Existing Flip 7 owners have far less reason to move. Last year’s model still delivers most of the same foldable experience, and its lower price makes the Flip 8’s refinements look expensive rather than essential.

Overall winner: Samsung Galaxy Z Flip 8

There’s no doubt that AI has helped revolutionize the world of work at many companies, with human employees working alongside agents to unlock improved efficiency and productivity.

But for many firms, the promise of AI is still being overshadowed by the sheer scale of the endeavor - with so many models (and new ones emerging on a seemingly weekly basis), how do you know what is right for you?

One of the big themes at the recent Atlassian Team Europe 26 conference was how companies can cut through the chaos and realize their aims, so I spoke to the company’s top AI minds to find out more.

The power of teams

“Atlassian's entire ethos is all around teams - unleashing the power of the team is our mission,” Atlassian Chief Product & AI Officer, Tamar Yehoshua tells me at Team Europe 26.

“We've all been using AI agents (but) they're all more geared towards an individual,” she adds, “but they're not organized or effective for teams...so we're taking that and saying, ‘how do you work effectively with teams and AI’? That's multiplayer, and so we want to make sure everyone can work effectively, so you can collaborate with agents in an effective way like you can collaborate with humans.”

This “multiplayer” AI was another major theme at Team Europe 26, as Atlassian looked to emphasize its position as being the connector between customers and technology.

“Teamwork is very messy, and our job is to make it much less messy, and understandable and controllable for our customers,” Atlassian Head of AI Sherif Mansour says.

“A lot of AI is about increasing personal productivity…the same problems we solve for humans, we solve for agents.”

(Image credit: Atlassian)

“Every customer we talk to, everyone is using AI - it's no longer should we, or what do we do - that's already happened...but people are not sure how to use it effectively, and they're not sure if they're getting the maximum ROI,” Yehoshua adds.

“You see the evolution with every new technology, first there’s scepticism, then curiosity, and then its acceptance, and then it’s optimising - I think we’ve gotten through the initial phases, and now we’re asking, how do I use this effectively and optimize it?”

She notes there are different levels of anxiety for customers, as the sheer number of new AI models coming out can be disorientating, especially with the need to stay flexible for when new models launch.

She notes that companies don’t want to be using advanced models for simple tasks - something Mansour refers to as “killing a mosquito with a rocket launcher” - so making sure the right model for the job is on offer is vital.

Atlassian’s AI gateway looks to pick the best model for the task within its products - so for example, Rovo work will switch to a new model if a more advanced one is released, all without the customer ever knowing - although they retain the ability to pick their own model if they want.

Yehoshua also highlights new upgrades to Loom, which now offers video generation from pull requests, and AI agents which use Rovo to categorize defects and cut down on duplicate reports.

No silver bullets

“AI feels like chaos to many of us - but I don’t think the answer is giving them a silver bullet that will solve all their problems,” Mansour says.

“We like to use the phrase ‘give them a nervous system’ so they can handle it - we’re always thinking about how we can bring calm and visibility and control to the chaos.”

Mansour says he advises customers not to bet entirely on one provider for the whole context graph, as the real value comes from collaborating with multiple entities to gain different kinds of opportunities and insights.

“There’s a lot of silver bullet offerings…but no-one really talks about the messiness of work,” he says, “it’s about creating a system to help you control the chaos - the chaos will always be there, but how do you make sure you’re aware of it?”

“It’s less about making teamwork perfect - it’s about making teamwork less messy.”

(Image credit: Atlassian)

I ask Yehoshua about personal usage - is she embracing the chaos of AI, or taming it for maximum efficiency.

“Everything that I do, I ask myself first - if this is a process I could change?,” she says, “if you look at AI and how it’s working, the biggest thing is behavior change…I tell customers, there is so much power in the tools they already have, and building in that behavior change is important, but it has to come from the top-down.”

How do we make sure we don’t go too far, and just automate all work processes, I ask her.

“I think it’s going to be trial and error - I think we’re going to try things and some of them might not work - but I honestly don’t think that’s our problem yet,” she replies, noting that enterprises often have the guardrails and protections needed to ensure a human always remains in the loop, and the answers can be trusted.

In an earlier session, Atlassian CEO Mike Cannon-Brookes had declared that, “customers are not subscribing for the software we've built, but for the software we will build”.

Normally you might question such a forward-looking statement, but with Atlassian’s focus on collaboration, it seems to ring true.

“The pace of delivery has increased significantly,” Yehoshua says, “you want to align with the companies that have the same vision as you do…people in some ways want more flexibility, as things are changing so fast, especially with models, they don’t want to be locked in…we are a neutral party who want to make it easy for our customers to work - we’ve always been the open platform who focuses on the team.”

“The main advantage we have is our platform,” she adds, “the strength of what we do is that customers are using us across so many different areas of their workloads.”

Do more - with more

Looking forward, it’s clear AI is here to stay - and it’s down to businesses themselves to judge how best to embrace it.

“I think we've now empowered more people to do more,” Yehoshua says. “Anyone can build a prototype - now that doesn’t mean that you want anybody checking in production code - so you still have to have very clear guidelines on what can go into production - but the idea is that it’s democratized, where the ideas can come from…so you have so many more options on how the lifecycle can evolve.”

“Worlds are expanding - they’re not going to be merged into one, but everyone can do more…every role can do more, so that everything is accelerated.”

Mansour agrees, pointing out that harnessing the power of context in your data can be the ultimate differentiator between your business and the competition.

“If we all have access to the same intelligence, that’s like a level playing field, it’s the sum of all averages - what’s unique is the actual perspective you bring to the table,” he says.

“Our customers shine when their context shines - and so anything we can do to help them enrich the context, understand it better to provide it to AI in a secure way, they will always get better results from that.”

• Stream The Party free on ITVX (UK restricted)
• Unblock ITVX with NordVPN (75% off now)
• Watch Episodes 1-5 now; Sunday, October 11

ITV’s new thriller The Party stars Luke Evans as journalist Martin Gilmour and Tom Cullen as his childhood friend Ben Fitzmaurice.

Based on Elizabeth Day’s bestselling novel and adapted by Sarah Solemani, the five-part drama follows a celebration that takes a shocking turn. The Party also stars Joanna Scanlan and Lydia Leonard.

So how can you watch The Party on ITVX from outside the UK? Can you get a free stream in the US, Australia or Canada? And is ITVX available as a smartphone app?

Here's our full (and quick) guide to how to watch The Party for free...

How to watch The Party for free on ITVX

Watch all five episodes of The Party for free on ITVX from Sunday, October 11.

You don't need a ITVX subscription, just a valid TV license and to sign up with your email.

New to ITVX? Register for a free account or download the app on your mobile device.

OUTSIDE THE U.K.? ACCESS ITVX FROM ANYWHERE WITH A VPN.

How to watch The Party on ITVX from anywhere

ITVX is only available to UK residents with a TV license signed up to the free service.

Luke Evans fans traveling or working outside the UK will need to use a VPN to access ITVX's free The Party episodes in the US and Australia.

There are lots of VPNs but NordVPN is the one you can rely on to unblock ITVX and stream ITV thriller The Party (2026).

NordVPN – try it risk-free for 30 days
Looking to access ITVX from outside the UK? If you're traveling abroad, use NordVPN to access ITVX as if you right back home in the United Kingdom. We test all the providers and we rate Nord as the best VPN. There's 24/7 support available, a money-back guarantee and, best of all, you'll currently get over 75% off with this deal.View Deal

It's really easy to use a VPN to watch The Party on ITVX.

1. Install the VPN of your choice. As we've said, NordVPN is the best choice.

2. Choose the location you wish to connect to in the VPN app. For instance, if you're visiting the US and want to watch your free usual ITVX stream, you'd select 'United Kingdom'.

3. Sit back and enjoy the action. Head to ITVX create an account or sign in, and watch The Party free.

Which devices can I watch The Party on ITVX with?

You can use ITVX on all of the following devices and platforms:

Amazon Fire (Tablets, Cube, Stick, TVs)
Android TV (please note: some models aren’t supported)
Android (Mobile & Tablet) - Android 7.0 and above
Apple TV (tvOS 14 or later)
Google TV (Chromecast with Google TV and NVIDIA Shield)
Freely
Freesat (please note: some models aren’t supported)
Freeview Play (TVs and set-top boxes)
iOS (iPhone & iPad) - iOS 14 and above
LG Smart TVs (2016-2024)
NOW Smart Sticks and Boxes (minimum firmware v11.5.0)
PlayStation (PS4 and PS5)
Roku (Stick & Roku-OS powered TVs, minimum firmware v11.5.0)
Samsung Smart TVs (2017 and above)
Sky Q, Sky Glass and Sky Stream puckVirgin Media (360, Stream, TiVo)
YouView (BT, Humax, Sony, TalkTalk)
Xbox (One, Series X, Series S)

The Party (2026) – Full Cast

Luke Evans as Martin Gilmour

Tom Cullen as Ben Fitzmaurice

Sarah Solemani as Lucy Gilmour, Martin's wife

Joanna Scanlan as Sylvia Gilmour, Martin's mother

Lydia Leonard as Serena Fitzmaurice, Ben's wife

Lindsay Duncan as Lady Katherine Fitzmaurice, Ben's mother

Douglas Hodge as Lord George Fitzmaurice, Ben's father

Sam Troughton — role not specified

Sally Scott — role not specified

Sinéad Matthews — role not specified

Bally Gill — role not specified

Amaka Okafor — role not specified

Rakie Ayola — role not specified

Also listed in the cast are Phoenix Laroche as Barnaby Fitzmaurice and Tilly Walker as Cosima Fitzmaurice.

Martin vs Ben: We break down the main characters...

Luke Evans in The Party (Image credit: ITV)

Martin Gilmour (Luke Evans)
A calculated and buttoned-up man, Martin is a political journalist whose opinion pieces always seem to favour Ben Fitzmaurice’s position. As the new scholarship boy from South Wales, Martin fell in love with Ben at school and has kept himself in Ben’s circle ever since, by whatever means necessary.

Martin remains loyal to Ben, despite a secret in their past which caused a seismic shift in their relationship, and these days has to contend with Ben’s growing family and political ambitions. That is until the secrets that have bound Martin to Ben threaten to come out…

Ben Fitzmaurice (Tom Cullen)
A confident, charismatic man who has had the world laid at his feet. Born into landed gentry, Ben is a Conservative Party MP with eyes on the party leadership. He has a charm that comes with growing up with an innate safety net.

A gifted speaker, Ben has the ability to hold a room with ease, but beneath this bluster run deeper emotions that Ben has yet to face up to. When his own impulsive decisions throw everything into disarray at his birthday party, Ben is forced to confront his dark past head on. As the hedonism of the party rises the temperature, Ben has to reckon with the truth of what he did… can he really run from the truth forever?

The Party launches on Sunday 11th October with Episode 2 airing on Monday 12th October, on ITV1, ITVX, STV & STV Player

We test and review VPN services in the context of legal recreational uses. For example:1. Accessing a service from another country (subject to the terms and conditions of that service).2. Protecting your online security and strengthening your online privacy when abroad.We do not support or condone the illegal or malicious use of VPN services. Consuming pirated content that is paid-for is neither endorsed nor approved by Future Publishing.

Solar power continues to be used more widely, and the science behind the technology is developing too: researchers from the Institut de Ciencies Fotoniques (ICFO) in Barcelona, Spain have just announced a breakthrough in solar cell engineering that could push efficiency rates beyond the standard 33% ceiling.

When the solar panels of today convert sunlight to electricity, most of the energy is lost along the way, through heat and light — one way to think about it is as the 'cost' of conversion. A physics theory known as the Shockley-Queisser limit predicts that we'll only ever be able to hit a 33% conversion rate in standard single-layer solar cells.

However, scientists like the team at ICFO are hard at work trying to break the Shockley-Queisser rules. In this particular study, the focus is on the faint infrared glow that solar cells give off: it's technically called the Boltzmann loss, and it's part of the reason today's solar tech is stuck below that 33% limit.

Beat the glow

The ICFO research team (Image credit: ICFO)

When sunlight hits a solar panel, the photons in the light knock electrons loose in the panel's semiconductor material. Those electrons can be pushed out as electricity, but not all of the electrons are collected in time: some escape and are turned back into photons, which creates an infrared glow spilling out in all directions.

To tackle the problem, the ICFO team took a standard solar cell and swapped out the transparent front electrode — that's the bit the sunlight strikes first, and it has to both conduct electricity while also letting as many photons through as possible.

Rather than using the standard indium tin oxide for a front electrode, the scientists used a very thin layer of silver, thin enough to let most of the light through, operate as a conductor, and also behave partly like a mirror. This restricted the angles through which light could escape, boosting the efficiency rate.

"Our study is a significant step forward in the solar energy conversion field as it opens a completely unexplored route to surpass the Shockley-Queisser limit," says Dr. Francisco Bernal Texca, the first author of the new study.

By adapting and evolving the approach used here, the scientists say that efficiencies close to 43% could be possible, in theory — though it's important to note that the team only observed small gains in efficiency in their tests. The 43% claim is a prediction about the potential of this method.

This 'mirrored box' approach to solar cells has been explored in the past, but no one's been able to get it working in a lab before, or to get boosts in both voltage and current at the same time (a critical part of the setup).

Stacking and splitting

Longi has had an efficiency announcement of its own to make (Image credit: Longi)

As the ICFO researchers point out, there are multiple angles to approach solar panel efficiency from, and there's progress being made in all of them. There are basically three ways that energy is lost from sunlight in a solar panel: through light loss (as above), through heat loss, and through the photons having too little energy to be converted.

There's a lot of attention being paid to the heat loss part of the problem, and the most common way scientists are trying to get around this is by stacking two materials together. These have different bandgaps, which are essentially the minimum energy a photon needs to kick an electron loose. A bigger bandgap means more photons can be caught (it captures a greater spectrum of light, essentially).

Earlier this year, Chinese solar manufacturer Longi set a new record of 35.5% efficiency by stacking a crystal material called perovskite on top of a conventional silicon cell. The bandgap of perovskite can be 'tuned' to accept specific wavelengths of light, which means higher energy photons can be absorbed.

You'll see a lot of different records and percentages announced though, which relate to different materials and different solar cell technologies. For example, researchers from Kyushu University and Johannes Gutenberg University Mainz recently announced a quantum yield of 130% with their own approach, using singlet fission.

That's a yield though, not an efficiency rate: to tackle solar panel heat loss, singlet fission splits high energy photons into two packets, which can then free two electrons, and in this case it produced 13 energy packets for every 10 photons. It could eventually improve solar cell efficiency further down the line, but more work is needed to develop it.

While this solar panel science can get quite dense and complicated, the end goal is simple to understand. As these technologies are refined and commercialized, we'll be able to get more electricity from solar panels without increasing their size, cutting utility bills and reducing our reliance on fossil fuel energy.

Conrad's (Pierce Brosnan) quietly disarming when he wants to be, isn't he? As MobLand season 2 hits the halfway mark, his power play on Eddie (Anson Boone) confirms that his priorities are in completely the wrong place.

Not only is he burying his head in the sand when it comes to the impact of his civil war feud with Kevin (Paddy Considine), but he's also neglecting his health in a way that could have serious consequences.

But what on Earth will happen this week? And when does MobLand season 2 episode 5 arrive on Paramount+?

What time can I watch MobLand season 2 episode 5 on Paramount+?

MobLand season 2 episode 5 will drop on one of the world's best streaming services in the US and Canada on Friday, October 16 at 12am PT / 3am ET.

Here's when it will be released in other nations globally:

• US – 12am PT / 3am ET
• Canada – 12am PT / 3am ET
• UK – 8am BST
• India – 12:30pm IST
• Singapore – 3pm SGT
• Australia – 5pm AEST
• New Zealand – 7pm NZST

When do new episodes of MobLand season 2 come out?

(Image credit: Paramount+)

MobLand season 2 will have a total of 10 episodes, with new entries airing weekly. That gives us the following schedule:

• Episode 1: out now
• Episode 2: out now
• Episode 3: out now
• Episode 4: out now
• Episode 5: October 16
• Episode 6: October 23
• Episode 7: October 30
• Episode 8: November 6
• Episode 9: November 13
• Episode 10: November 20