OpenAI's debut of the new GPT-5.6 model prompted the usual ritual of benchmark charts and arguments over whether it is really smarter than the last version, but CEO Sam Altman asked for a little more this time. He publicly asked to see what people actually built with it.

That led to a much more interesting showcase. Developers responded with all kinds of ideas, pilot projects, and even complete services. It makes sense, since OpenAI claims GPT-5.6 is better at coding and more reliable for long tasks. Seeing them turn into real software says much more than a release blog ever could. Here are five that stood out among the deluge.

ChatGPT coworker

The demo from Kitsune Agent Lab almost makes the chat window feel old-fashioned. The AI agent is given a goal and gets on with the job, moving between different tools, making decisions, and keeping track of what it has already done.

The interesting part is how motivated the AI agent appears to keep going and how good it is at remembering what it's done before. Developers have been asking for something like this for a while. AI is far more useful when it can finish the work instead of simply suggesting how you might do it yourself.

Financial chatter

One of the most charming projects makes New York City look like it belonged inside an original Game Boy. It comes complete with chunky pixel graphics but runs on a live digital map of New York that pulls in real-time information, including subway trains, weather conditions, and ferry movements. Instead of wandering through a fictional RPG world, you're exploring a tiny, pixelated version of the city.

A project like this requires far more than generating a few lines of code. It brings together live data feeds, mapping, interface design, and plenty of problem-solving into something that feels polished rather than experimental. It's one reason developers are feeling excited about GPT-5.6

Wardrobe AI

This project uses GPT-5.6 to create a polished AI wardrobe assistant that organizes clothing, suggests outfits, and presents everything through an interface that feels more like a premium consumer app than an experimental AI demo.

It's an impressively complete experience. The application gives users a visual, interactive way to browse their clothes and receive recommendations based on what they already own. The demo also highlights GPT-5.6's strength in developers building entire applications instead of isolated features. It brings together interface design, image generation, organization, and intelligent recommendations that would normally require stitching together several complex systems. GPT-5.6 appears to handle much of that heavy lifting.

Pokémon Go, but for neighborhood cats

One developer made a whole real-world-based game called CatchCat. It's like a digital expansion to a scavenger hunt for cats. Point your phone at a real cat, let the app verify the sighting with its camera, and turn that encounter into a collectible digital cat card with its own personality, rarity, and place in your growing album. It is essentially a creature-collecting game in which the creatures are the neighborhood cats you meet.

Players can build collections, explore community sightings, compete with friends, and gradually fill a living scrapbook of feline encounters, all wrapped in a polished interface that would not look out of place on the App Store or Google Play. Building something like CatchCat means juggling computer vision, mobile development, backend services, and game design.

Tasteful travel

One of the most ambitious projects, Atlas Mode for Pearl, is an interactive globe that turns travel planning into something closer to exploring a living map. Instead of typing destination names into a search box, users can spin the globe, discover places visually, and receive recommendations for restaurants, hotels, and more matched to their personal tastes.

It has an impressive number of moving parts running behind the scenes. The app combines geographic data with an individual taste profile, then layers AI recommendations directly onto an interactive globe. It even has an audio aspect thanks to GPT Voice 2.1. You can talk through vacation ideas instead of endlessly tweaking search filters.

That is a recurring theme among the projects developers rushed to show Sam Altman. The AI is no longer the product itself. Increasingly, it is the engine quietly powering products that people might actually want to use.

The AI buildout is well and truly underway with OpenAI CEO Sam Altman leading the charge, having made various deals with companies including Oracle and Nvidia to guarantee the infrastructure needed to train future AI models is installed. But as this ensues, the spotlight has been thrown on how much energy these models will need.

Eating machines

Altman was speaking during an AI summit in India earlier this year when making the remarks to The Indian Express.

The OpenAI chief was being questioned about the substantial amount of energy that AI has already been consuming – and will be projected to continue consuming – for both training and inference. Bringing these models online, after all, and keeping them running require a huge amount of resources, not only in terms of the energy to power the data centers, but the water for cooling, and the components and resources in building the hardware.

Altman's defence hinged on the idea that people, too, require plenty of energy in order to reach utility — while AI can be trained much quicker. This view, in essence, frames machine intelligence as a like-for-like competitor with human intelligence.

Energy efficiency

Altman argued that humans are deeply inefficient, and that you should compare the total energy spent to create a human expert versus a machine expert. There's also an argument that energy efficiency of AI could improve over time.

But detractors were scathing in their criticism of this entire point of view – saying this framing doesn't take into account the fact that the human brain operates on roughly 20 watts of power. This isn't to mention the ethically gray and dehumanizing nature of the remarks.

OpenAI is building an AI-powered smart speaker nobody wants. That is, if you believe the Bloomberg report from Mark Gurman and you've read his description of said rumored device.

According to the report:

"OpenAI believes the product’s defining feature will be its personality and ability to connect on a humanlike level with users. The speaker incorporates mechanical elements that can move on their own, creating a sense that it is alive and not just an object responding to commands. The machine also will draw on personal information such as emails to better understand its owner."

The news sources, it appears, come from an insider who decided to spill all the juicy details mere hours after Apple dropped a blockbuster trade-secrets lawsuit on OpenAI's head. OpenAI claims it's done nothing of the sort, and recent reports say that Apple's claim that the AI giant has not even responded to Apple's earliest concerns was based on it potentially incorrectly identifying the former Apple employees who left to join OpenAI (allegedly with Apple trade secrets in tow).

Apple's concerns here are twofold: First, these former employees had access to many of Apple's secretive product development details and may even have asked recruits to share fresh details when they approached them to interview for jobs at OpenAI. The other concern is that Apple is already far behind in the AI race, and if Apple's plans for Siri, AI, and a potential robotic desktop home assistant were also leaked, it could harm its ability to catch up in multiple market sectors.

This latest news, which may or may not be accurate, should put Apple's fears to rest.

OpenAI is apparently not building something that could ably compete with any of Apple's key hardware or future hardware initiatives.

First of all, there's the smart-speaker-ness of the whole rumored OpenAI concept. There are already too many smart speakers on the market, many of them with their own smart assistants. Amazon, for instance, is smack in the middle of trying to convince millions of customers that not only do they need Echo devices throughout the home, but they need the AI-powered Alexa+ to guide them through their smart home experiences and, to some extent, their lives.

Apple has its own HomePod, Siri-infused speakers, which may get considerably more powerful with the Gemini foundation model-backed version arriving this Fall.

Put another way, smart speakers are a known quantity in the home consumer electronics space, and I think what most tech companies are realizing is that people like and use them, but mostly in limited ways: they want music, occasional answers to simple questions, and voice control of their smart home devices. That's it.

Why does my speaker think it's alive?

OpenAI appears to be prepared to offer something different: a personality-filled speaker that can watch you, move to engage, seem alive, and generally make you feel uncomfortable.

Obviously, that would not be the objective, but it could be the result. Who needs a speaker that quietly watches you as you walk from your kitchen to the den, waiting and hoping for you to say, "hey ChatGPT, what's up with the Strait of Hormuz today?"

In my home, we have a Psync smart security webcam with one oddball feature: it has a motorized body that can turn almost 360 degrees on its base and lift its thin, rectangular face and camera to keep track of people and alert me to intruders. However, most of the time, it's just watching us move around the house, and I can tell you that my family hates it. Sometimes I come home and find its face forced down so it can't pop up and track anything.

Now, imagine a larger and far smarter OpenAI AI smart speaker in your home, watching, waiting, chiming in when you don't want it to, and generally making people feel uncomfortable.

This will not be the breakout hardware hit OpenAI is hoping for.

Look, I was under the impression that OpenAI (really Jony Ive and Sam Altman) were working on an AI wearable. I didn't love that idea either, but it was a lot less creepy than this.

So, Apple, chill out. OpenAI's plans are no threat to you, even if they do allegedly have a bunch of insidery Apple information.

• AI hallucination can be weaponized, new report warns
• HalluSquatting is short for “adversarial hallucination squatting”
• GitHub Copilot, Gemini CLI, and OpenClaw are all affected

Your favorite AI service could be subverted to deploy code that turns your phone or PC into a botnet, according to researchers at Intuit, Technion, and Tel Aviv University.

The technique has been given the name HalluSquatting, a portmanteau of adversarial hallucination squatting, and is similar to typosquatting in that it relies on a mistake in order to distribute malicious code. While typosquatting might occur with the incorrect input of a website URL, HalluSquatting pivots on an LLM being unable to identify a resource or repository with 100% accuracy.

Relying on an LLM’s tendency to hallucinate repository resource identifiers, this weakness could be scaled up to conduct massive ransomware campaigns, botnets, and more.

Push-me-pull-you

Previous LLM-based malware operations have relied on pull-based attacks. In this scenario, a prompt designed to jailbreak or otherwise subvert the AI is (for example) placed on a website and the LLM encouraged to gather the information, thereby reducing its internal security.

What the researchers have shared in their paper, is that pull techniques are being combined with push attacks, which are traditionally executed as code injection.

The paper’s introduction summary states: “By preemptively registering hallucinated resources—a technique we call adversarial hallucination squatting (HalluSquatting)—we demonstrate remote tool execution and remote code execution at scale across a range of popular agentic LLM applications, which could be exploited to the establishment of a botnet.”

Once an attacker has identified the resource likely to be misnamed by an LLM, and squatted on it (to embed adversarial prompts), the work is done. All that remains is for a user to trigger the resource, the AI chatbot or agent to initiate the response, and the squatted resource will be accessed.

Promptware attack

Following this, the adversarial content held within the squatted resource is activated, triggering the tool invocation stage. This is the promptware attack, where attacker-controlled instructions are executed, with results potentially including turning the device you’re using into a botnet zombie.

LLMs such as the Cursor, Cursor CLI, Windsurf, GitHub Copilot, Cline coding assistants have been used in the testing of this avenue of attack along with Gemini CLI, and the OpenClaw, ZeroClaw, and NanoClaw AI assistants. The researchers successfully achieved remote tool execution (essentially remotely accessing and controlling the LLMs) and remote code execution (RCE, where malicious code is executed remotely).

Some mitigation is available, including LLM developers blocking fetch operations in favor of a search tool, and resource owners enforcing strict naming, perhaps in favor of globally unique resource names. However, these are will require collaboration by disparate parties, and may take a while to implement.

The risk of LLM-based malware is increasing, and some has already been spotted in the wild. Of these, the JADEPUFFER attack is perhaps the most notable, as it isn’t simply AI-based malware – it is a full ransomware attack run entirely by an LLM.

• The three man team is known as “Slopfix”
• It claims to be able to reduce AI generated codebases by up to 65% in size
• They aim to "refactor vibecoded codebases back to maintainability"

Vibecoding has a lot to answer for, not least some excessively large codebases. A new team of software engineers are collaborating to reduce the size of these cumbersome projects… with a $10,000 per week bill.

Slopfix is the name of the team (comprising a trio: Maciej, Kuba, and Krzysztof), but its aim is efficiency and functionality, rather than code golf, where code is reduced to the shortest possible length.

However, while this might seem like a noble task and a service worth paying for, Slopfix isn’t taking a stand against the use of AI. In fact, it is employing AI tools to detect the AI flab in your codebases.

Use an AI to catch an AI

Challenges around vibecoded projects have increased in recent months, as the limitations of the technology become apparent.

While using an AI to program based on your prompts and requirements is straightforward, agents habitually begin to lose context and logic once the project reaches a certain size or age. Once that happens, you’re looking at duplication, features breaking, and of course, the dreaded hallucination.

Slopfix is targeting companies that have adopted vibecoding, built huge codebases, and found that they’re running into issues. To find the problematic AI code, however, Slopfix is employing AI.

They state that a full “screen by screen, endpoint by endpoint” evaluation of the vibecoded app is made, which aims to find the duplicated functions, broken logic, and other issues. There’s also the promise of a two-week warranty for anything they break.

All of this is aided by Claude Code “on a very short leash” which Slopfix uses find problems. They clearly state that “the agent doesn’t get a vote.” Instead, they’re relying on their experience as developers to improve your code.

$10,000 seem a bit steep?

While the price might seem high, $10,000 for one successful week’s work for three seasoned developers shouldn’t really be a budget breaker.

The fee covers successful work only, and as the Slopfix website states, payment is in proportion to how much of the reduction target the team hits, with $10,000 being the price for hitting the target – it’s not the default fee.

However, there is a lot of preparation involved, and the analysis of your codebase is conducted free of charge. If they can't fix your project's issues, they'll let you know and refuse the contract.

As software consultancies go, Slopfix is an unusual case. But as the problems with vibecoded projects begin to become apparent, competing consultancies may begin offering similar services.

• UN Secretary General calls for a global ban on autonomous "killer robots”
• Guterres argues that delegating life-or-death decisions to machines is “morally repugnant”
• Governments should take a stance now – not wait for something catastrophic to happen

UN Secretary General António Guterres has called for lethal autonomous weapons, which he describes as ‘killer robots’ to be prohibited under international law following recent discussions at the first Global Dialogue on Artificial Intelligence Governance in Geneva.

Guterres’ demand to ban these weapons focuses on those capable of identifying, selecting and attacking targets without human oversight, which leaves artificial intelligence and other computer systems in charge of a life-or-death decision.

He ultimately argued that certain decisions must remain exclusively human, and the decision to take a life is well into the boundary of requiring human oversight. Transferring the decision-making to killer robots would be “morally repugnant” and “politically unacceptable,” he argued.

AI requires global regulation as military AI poses major threats

Key to the Secretary General’s argument is that he urges governments to take action and ban such robots now, rather than waiting for an autonomous weapon to cause a major incident before rethinking their strategies.

“Let us not wait for atrocity to act,” Guterres said. “Some decisions must remain forever human – none more than taking a human life.”

The issue is becoming more urgent now that AI models and advanced chips are already being used within military intelligence, targeting and other battlefield systems.

More broadly, Guterres’ thoughts align with those of Anthropic, which recently had a dispute with the Pentagon after seeking guarantees that its models would not be used for autonomous weapons or surveillance.

While the Pentagon had rejected those limitations, arguing that it should be able to use Anthropic’s models for any lawful purpose, the case highlights how private companies are becoming increasingly intertwined with digital warfare.

Reporting by the Wall Street Journal cited a similar view by Pope Leo XIV, who warns that AI-controlled weapons could promote an “anti-human” view of warfare. He warned that the autonomy could reduce some dangers and distance political leaders from the human consequences of conflict.

There’s a need to balance the pros and cons of AI

However, artificial intelligence does promise several benefits to modern warfare, particularly in its ability to process huge amounts of information extremely quickly. With modern compute, militaries can respond to threats at lightning speed, improve their accuracy and precision, reduce soldier risk and potentially reduce civilian casualties, too.

Critics also question whether human oversight of AI systems is at all meaningful if the person in charge only has seconds to act on AI-generated information in the first place.

It’s also yet to be determined which party or group of parties should be held accountable for any incidents or mishaps – human operators, commanders, hardware manufacturers and software developers are just some of the parties up for judgment.

“We may be the last generation able to set the terms on which humanity and machines coexist,” Guterres warned separately in an X post, warning that AI must be governed, trusted and fair.

“It sounds like science fiction, but it's a real possibility, and it could change the world in ways that we don't understand yet, and it could change the power dynamics of our planet in ways that require our attention,” Independent International Scientific Panel on AI Co-Chair Yoshua Bengio added.

It all started with an email. Someone was reminding me about the fast-approaching National AI Day. I never finished the email but quickly wondered, "National AI Day, is that really a thing?" It is, and I found the man responsible for it.

"We have about 30,000 applications a year for new national days. From that, we have a committee of people look at them," said National Day Calendar.com founder Marlo Anderson when I reached him by phone just two days before the big July 16th event.

National AI Day, though, did not follow that traditional route when it was designated a day on their calendar in 2025. Anderson told me he does a lot of AI work with the business and has been using AI for 20 years, so he did something slightly unusual.

"We just figured having an AI day would be appropriate," he said and admitted he did the designating.

It is not the normal process, though Anderson believes, but could not specifically recall, that people had suggested the day in the past.

You made this list

It is a big deal to get added. After all, National Day Calendar only adds a handful each year and, as Anderson admitted, some already believe they have too many "National Days" (he noted that he might be among that cohort, too).

Still, Anderson, who founded National Calendar Day in 2013, believes AI warranted it. He explained that it's already responsible for much of the "mundane process" work they do, including uploading the National Day Calendar's daily videos to a platform called Video Elephant. They were uploading 100 clips as he spoke to me. Doing it by hand would take two weeks. With AI, "just one day or two."

As for which AI platforms Anderson uses, he seems to spread it around, telling me that Claude is used for app development and website maintenance, ChatGPT Voice for brainstorming projects, and Gemini in National Day Calendar offices in North Dakota.

Anderson is a firm believer in the power and potential of AI for both his own work and as an agent for good in the world. Locally, he told me, "If we have an agentic AI that can handle that workload for us, we should probably do it."

National Anti-AI Day

When I asked abotu the AI backlash — a recent study found 40% of surveyed peopel are limiting use of AI — Anderson told me, "We talk about it all the time," but added, "We get backlash on National Cheese Sandwich Day and French Fry Day," which is to say, he's not sure if the backlash is any greater than for other oddball National Days.

More seriously, Anderson is well aware of the debate about AI's impact on the environment but is also convinced the positives far outweigh the negatives.

"I also understand there’s a lot of benefit," like the ability to find medical cures. They're making, he told me, "remarkable progress right now" in medicine.

He then spun out an analogy about horses and cars. A million hours ago (in 1912), most people were still riding horses and, it turns out, they were pretty dangerous, too. Early cars, with the lack of traffic infrastructure, weren't much better, but "most people would agree a car is a better way to travel," he said, adding that we're currently "at the same crossroads with Artificial Intelligence."

Ultimately, Anderson's decision to create National AI Day was rooted in its current impact.

In 2025, "the conversation had heightened to a point where probably everyone knows about AI, at least in the US. Everyone is probably using it, whether they know it or not," he told me.

It's also about AI's long-term prospects.

When goat yoga was a big thing, Anderson explained, they had a lot of requests to make it a day. "There's no day because we assumed it would be a fad [he was quick to add he’s sure it’s wonderful]. 20 years from now, no one will know or talk about goat yoga, but 20 years from now, they will be talking about AI."

So happy National AI Day. Feel free to celebrate on July 16 by using it, deriding it, or ignoring it altogether.

• Microsoft CEO Satya Nadella has warned AI companies are training their models on the business secrets of their customers
• These secrets are then used to train new, more powerful models, that are sold to their customer's competitors
• But, Nadella says there is a way to remain competitive without being locked in to one AI vendor

Microsoft CEO Satya Nadella has warned the big players in the AI industry are using their proprietary models to learn the business secrets of their customers, which they can then use to train and deploy more advanced AI models.

The crux of the issue, Nadella said in a blog post, is that, “You essentially pay for intelligence twice, once with money, and again with something even more valuable: the proprietary knowledge you must reveal to make that intelligence useful. The better you want the model to perform, the more of that knowledge you have to feed it!”

What Nadella is saying in essence, is that AI companies are harvesting sensitive business data from their customers, using it to make training their models cheaper, and then launching these models for use by their own customer’s competition.

“The kind of knowledge a competitor could never buy”

“Models learn from ‘exhaust,’ the prompts people write, the tools agents use, and especially the corrections people make when the model is wrong. Every correction is distilled into institutional know-how,” Nadella explained.

Nadella also criticized how AI companies are increasingly complaining about how their models are being distilled by their own competition. For example, Anthropic accused retailer and e-commerce company Alibaba for using thousands of Claude prompts to distill their own models. By figuring out how a proprietary model works, you don’t have to spend the enormous amount of capital needed to source training data and create your own AI model.

This, for Nadella, is a major contradiction in how AI companies work. “While the great innovation that comes from model providers having fair use rights to train models on public data is needed, I find it ironic that the status quo is to then turn around and impose restrictive terms on distillation,” he said.

It is also therefore hypocritical for AI companies to accuse other companies of distilling their own product, and then include within their AI usage contracts clauses that allow AI companies to “reserve the right to learn from customer usage and interaction data.”

“In consuming intelligence, you are creating intelligence. And what you create should belong to you,” Nadella added.

On-prem is back in fashion

Nadella’s fix for this growing problem? It’s time to move back to on-prem. Nadella encourages businesses to “retain ownership” of the data they feed AI models by switching to the use of “proprietary learning environments” built on the cloud.

The added benefit of moving to these environments is that they allow businesses to switch between different AI models provided by different companies using “orchestration layers” and AI gateways.

There is also a growing trend of businesses switching to using open source technologies, which goes hand in hand with businesses operating in the cloud. Businesses can train open source AI models using their data that is already available in cloud environments to do much of what the proprietary models do, for far cheaper — and without handing over that same sensitive data to be used by AI companies to train their own models.

The on-prem solution also has additional benefits. AI models operated on-site within manufacturing plants, stores, and other premises are far cheaper and require less specialized hardware. Businesses that operate using a centralized cloud are increasingly encountering issues with data egress fees, storage bloat, and idle specialized hardware.

Google Cloud recently released a report about these very issues, and also encouraged businesses to move towards using AI gateways and on-prem models to reduce latency, improve resilience, and cut per-token costs by switching to local, highly optimized models.

Via TechCrunch

• Study finds social media posts are increasingly AI-generated
• LinkedIn particularly affected, with 40% of long-form posts written by AI
• Substack and Twitter/X also badly hit

LinkedIn and other social media networks are rapidly being consumed by AI-written slop posts, new research has claimed.

A report from AI detection firm Pangram Labs found nearly half of all long posts (over 250 words) on LinkedIn were created entirely by AI, with the likes of Substack and X/Twitter also seeing a huge rise in such content.

"LinkedIn was the most AI-saturated platform, where more than 40% of longform posts flagged as fully AI-generated," the company's report said.

Social media drowning in AI slop

The study, which also examined Medium and Reddit alongside the other social networks for a data set of over a million posts, found one in four longform posts on social media flagged as fully AI-generated, with lengthier content much more likely to be created with AI than shortform.

Pangram found that LinkedIn was the most AI-saturated platform, where more than 40% of longform posts were flagged as being fully AI-generated, with Substack the least affected, with longer posts often far less likely to be AI-generated.

LinkedIn was also identified as having the highest AI share of any platform included in the report, as although its posts only made up a third of scanned items, it accounted for nearly two-thirds (62%) of all AI content flagged by the system.

"Professionals come to LinkedIn to hear from real people and their unique insights and perspectives," a LinkedIn spokesperson told us.

"We actively work to reduce low quality, automated or generic content, and while AI can be used to beat the blank page problem, our focus is on surfacing professional conversations that help people advance their careers. You can learn more about how we’re keeping the Feed trusted and professional here”.

However, when mixed AI and human content were included, X/Twitter was by the most swamped by AI, with the study finding almost half of articles on the site were either fully AI-generated (23.9%) or AI-assisted/mixed (22.9%), with only 53.2% of X articles flagging as fully human-authored.

"Our data shows that AI-generated content is a problem across all platforms, and it is hitting longform content especially hard," Pangram said.

"Contrary to what one might expect, people are overwhelmingly willing to use AI to speak on their behalf in professional settings that are associated with their real identity, and less likely to use it on casual and anonymous platforms."

"AI writing is now a problem everywhere on social media," Pangram Labs CEO and co-founder Max Spero noted. "An internet that is completely flooded with undisclosed AI content is bleak, but we don't believe it's inevitable."

• Americans want the massive wealth of AI firms added to a sovereign wealth fund
• 69% would see AI firms forced to transfer 50% of stock into a sovereign fund
• The fund would help redistribute wealth and back new infrastructure and developments for working class Americans

A national survey has found over two-thirds (69%) of US citizens want to see AI firms transfer half of their stock into a sovereign wealth fund.

The survey, conducted by Verasight among 1,690 adults, also found that there was overwhelming support (89%) for AI companies to publicly disclose the results of all internal safety testing.

The sovereign wealth fund, proposed by Senator Bernie Sanders, would provide wealth for current and future generations, as well as acting as a source of capital for investment in new projects and developments designed to improve the lives of working class Americans.

Americans want AI wealth redistribution

At the announcement of Sanders’ proposed American AI Sovereign Wealth Fund Act, the senator said, “It would guarantee that the economic benefits generated by AI are used to improve the lives of all of us — not simply to make the richest people in the world even richer.”

“The future of AI and the fate of humanity must not be decided behind closed doors in Silicon Valley by billionaires seeking to maximize their power and profit,” Sanders said.

Interestingly, the survey only saw a small dip in support to 64% when the sovereign wealth fund was tied directly to Sanders, showing the bi-partisan desire for the enormous growth in AI wealth to be redistributed among Americans.

According to a Goldman Sachs report, companies operating in the AI industry have added more than $27 trillion in market value since late 2022, with corporate profits and tech investment soaring. But at the other end of the scale, working class Americans are seeing jobs replaced and entry level positions disappearing due to AI technologies.

A further report from Goldman Sachs predicts that during the 10 year AI transitional period, up to 15 million US workers could lose their jobs - around 9% of the current US workforce.

In 2026 alone, the tech sector has seen more than 166,000 layoffs, with many attributed to the adoption of new AI technologies. The trueup layoff tracker expects this number to rise to 312,000 by the end of the year.

Electricity prices are also surging in the US due to the demand of AI data centers, raising the cost of day-to-day life of millions of Americans. As a result, US representatives have put forward a bi-partisan Ratepayer Protection Act that would force AI companies and hyperscalers to pay for the energy they use, with additional charges to help fund the expansion of electricity infrastructure that has been placed under additional load by data centers.

Via CNBC

• Windows 11's Copilot app has a new feature in testing
• 'PC insights' provides an easy way to receive clear answers to hardware-based questions about your device and its specs
• While there are some fears over privacy (and bloat), Microsoft has made it clear that Copilot needs to be granted permission to access your system and files

Copilot is getting a new ability to answer questions about your PC's hardware, allowing the AI to tap into the relevant hardware details to do so – and while Microsoft is treading carefully with privacy here, that's unlikely to stop some level of paranoia.

Windows Latest flagged the introduction of 'PC insights' for the Copilot app on Windows 11, which as Microsoft explains, "enables customers to conversationally ask Copilot questions about their Windows PC and receive clear responses based on their device's state without having to dig through system settings."

This is currently an experimental feature, so still in testing, and an optional ability that you must turn on for it to be in play. Windows Latest notes that it's gradually rolling out, but only in the US for now.

You can ask Copilot how much RAM you have, or storage space left, or what your GPU is, and the current level of usage for your processor, and a whole bunch of similar component-related queries. You can ask about elements as diverse as whether you have an antivirus running, or what your laptop's battery health is, diving into mild troubleshooting territory should you wish.

To get its answers, the Copilot app hooks up to Windows APIs to analyze your system, and the AI asks for permission to do this. You can allow it access to your PC's hardware details on a one-time basis for that session only, or you can elect to 'always allow' if you're happy to give Copilot this access on a more permanent basis.

Analysis: fears over hallucinations and bloat

As ever, this is AI, and as Microsoft notes, Copilot "may not always provide complete or accurate information", especially during this testing phase. So, if you do get a chance to try out PC insights, maintain a healthy sense of skepticism with the responses you get.

As Windows Latest makes clear, there's also a certain irony about a Windows 11 user checking up on resource usage, perhaps due to system sluggishness, employing the Copilot AI to run diagnostics when the app itself uses the best part of 1GB of RAM when running in the background and doing nothing.

That doesn't stop this new PC insights feature from being situationally useful, of course. Some of the reaction has come from a place of disdain, though, as you might guess, with comments such as the one from this Redditor: "Oh hey it's like Task Manager except instead of lightweight and authoritative, it's bloated and might be lying to me."

Of course, this is a feature aimed at less well-informed PC owners, not those who can easily understand what's happening in Task Manager at a glance. Criticism around the bloat of the Copilot app is fair enough, mind, and this is because in its most recent incarnation, Microsoft changed things so the app is essentially a standalone spin-off of the Edge browser.

Another worry is that of privacy, and having Copilot 'snoop around' on your machine, but as noted, there are clear requests for permissions, and the new feature is strictly opt-in. You don't ever have to go near PC insights if you don't want to. It's also worth noting that giving the Copilot app access permissions doesn't mean it can read the actual contents of files, but only their sizes (for weighing up questions about storage and the like).

At the moment, this is a purely informative or troubleshooting feature, and in the case of attempted diagnostics, it may point to issues with your PC, but won't resolve them for you. However, it's not difficult to envision where Microsoft might head with this, in terms of getting Copilot to implement fixes for certain issues that the AI flags up. I'm talking simple Settings changes rather than anything in-depth, and this has always been the idea of Copilot (even though it hasn't yet been realized to much of an extent).

When we get AI agents in Windows 11 – and they are coming, make no mistake – this kind of functionality may turn into a full-on troubleshooting agent. The trouble (pun not intended) with that being that the mistakes and hallucinations that AI can make could be considerably more aggravating in this kind of scenario.

TeraBox built its name on 1TB of free cloud storage, which is still the headline pitch. What's changed is everything sitting next to the storage. An AI Presentation Maker, an essay writer, a scanner, a transcriber, and a research tool called Deep Research are now baked into the same free account I've used for years.

Until recently, TeraBox functioned more like a toolkit than a dedicated AI assistant. Many of the tools only did one job, like generate a deck, paraphrase a paragraph, or scan a document — then handed back results rather than a conversation.

But now, Terabox has added a dedicated research assistant that can access the web and parse through complex queries for information. It can also generate graphics and create properly-formatted documents to help with your research tasks. It's neat, but the most capable versions of these tools sit behind a Premium+ subscription priced under $4 a month.

I've covered hosting, storage, and AI software for TechRadar Pro since 2012, including our 2026 buying guide for vibe coders. If you'd like to see a wider selection of dedicated AI tools to pick from, check out our list of 70+ other AI tools across different categories. For this Terabox AI review, I had access to a 7-day trial of Premium+, which let me try all of the platform's features in depth over the review period.

What is TeraBox AI?

TeraBox AI is the set of generative tools Flextech Inc. has added to TeraBox, the free storage app it took over from Baidu in 2020. Rather than one assistant, it's a handful of single-purpose tools, an essay writer, presentation maker, paraphraser, transcriber, scanner, and research assistant, living inside the app I use for backups.

I'd call it a productivity add-on rather than an AI platform in its own right. You pick a tool, type or upload your input, and get a finished result. That suits someone wanting a quick deck or a tidier paragraph, but the platform left something to be desired when it came to more complex tasks requiring advanced reasoning.

TeraBox AI: At a glance

Buy it if…

• You already use TeraBox for storage. The AI suite rides on cloud space you may already pay for, so the extra cost is small.
• You need quick scans and transcripts. AI Scan and AI Transcribe handle everyday OCR and meeting notes well for casual use.
• You want a fast presentation starter. One prompt turns into a usable slide deck in seconds, ready for further editing.

Don't buy it if…

• You need a serious writing or design tool. Dedicated tools go deeper than TeraBox's essay writer and paraphraser for professional output.
• Data jurisdiction matters to you. TeraBox's storage business began inside Baidu before Flextech took over in 2020, which still gives some professionals pause.
• You can't stand ads. The free tier's AI tools sit behind the same ad-supported experience as TeraBox's storage.

My time with TeraBox AI

I tested TeraBox AI through the web app and Android client, running presentations, transcriptions, and scans over several days. The Presentation Maker stood out. A one-line prompt about small business marketing trends produced a ten-slide deck, icons and chart included, in under a minute.

AI Scan and AI Transcribe felt the most useful day to day, turning a printed invoice into clean, editable text in seconds. Transcribing a short interview gave me a readable summary alongside the full transcript. The essay writer and paraphraser, by contrast, were serviceable but generic.

However, my experiences with the dedicated AI chat and Research Assistant were somewhat mixed. A simple query asking for Elon Musk's updated net worth returned accurate input, but a more complex task involving generating a graphic of his net worth over the years seemed to fail entirely.

TeraBox was able to offer the information as a table instead on further attempts, which led me to believe that the failure was a result of being unable to pull up the necessary integrations for data visualization rather than an error in the research itself. That does redeem it in my eyes given the price point, just don't go in expecting the same level of functionality as a dedicated AI platform like ChatGPT.

On value, Premium+ is an easy call. It costs under $4 a month and bundles 2TB of storage with the full AI suite. The caveat is that nothing here matches a tool built solely around AI writing or design.

TeraBox AI: Features

TeraBox AI isn't the best do-everything assistant, which also shows in how the features are organized. Presentation, writing, scanning, and research tools each live in their own corner of the AI tab, which keeps things simple but means there's no unified chat for mixing tasks together.

The Presentation Maker impressed me most, with Agent Mode generating decks up to 40 slides from a prompt or uploaded document, complete with citations when it pulls from web sources. A separate Beautify option restyles slides you've already made. For students or small business owners needing a deck fast, this alone might justify upgrading.

AI Scan and AI Transcribe are the most practically useful tools for office work, handling OCR, ID document capture, and on-the-fly translation on one side, and audio-to-text conversion with an AI summary on the other. Both worked reliably in my testing, needing only minor cleanup afterward.

Deep Research, the suite's research tool, builds a structured outline you can edit before it writes a full report. The output reads more like a market briefing than original analysis, so treat it as a starting point rather than a finished document.

The essay writer and Smart Paraphraser are the weakest links. Output reads competently but generically, and TeraBox doesn't disclose which model is doing the writing. Given the price for Premium+, the overall feature set still feels reasonably generous for a bundle riding on storage you might already want.

TeraBox AI: User experience

I got started in under a minute. I logged into my existing account, tapped the AI tab, and was generating a presentation within seconds, with no separate sign-up or onboarding flow to slow things down. Each tool opens to a simple prompt box, so the learning curve is close to zero.

But the experience feels bolted onto a storage app rather than designed around AI from the ground up. Switching tools means backing out to the main AI menu each time, and the free tier's ads occasionally interrupt what otherwise feels quick and uncluttered.

TeraBox AI: Customer support

Support looks decent on paper. Users across review platforms rate TeraBox's customer service 4.0+ out of 5 and Flextech representatives respond to public reviews directly, including the ones flagging slow replies.

I didn't need to contact support during testing, since the AI tools worked as expected. The bigger caveat is documentation, since TeraBox's help content for the AI suite leans on blog posts and FAQs rather than a structured knowledge base.

TeraBox AI: Pricing

• Free tier includes limited AI access, plus 1TB of ad-supported storage.
• Premium adds 2TB of storage and removes ads, but doesn't unlock the AI suite.
• Premium+ is the only tier with full AI access, priced from around $3.89 a month or $39.99 a year.

TeraBox's free plan gives a limited taste of the AI tools, including a small number of free presentation generations, on top of the full 1TB allowance. That's enough to test whether the tools suit your workflow before paying anything.

Premium+ is where the AI suite lives, bundling the essay writer, presentation maker, transcriber, scanner, paraphraser, and search tools alongside 2TB of storage for under $4 a month. TeraBox's official pricing page renders dynamically and didn't return visible data when I checked the source directly, so these figures come from cross-referenced third-party listings instead. There's no separate API pricing, since TeraBox doesn't offer developer access to the AI tools.

TeraBox AI: alternatives you should consider

• pCloud: A privacy-focused cloud storage service with optional zero-knowledge encryption, starting around $49.99 a year for 500GB, without TeraBox's bundled AI tools.
• Google Drive with Google AI Pro: Google's storage plans now fold Gemini directly into Drive, Gmail, and Docs, bundling 5TB of storage with AI access for $19.99 a month.
• Canva: A more polished option built for AI-assisted presentations and design, with Magic Studio included in Canva Pro for around $15 a month.

How I tested TeraBox AI

• Used the Presentation Maker, AI Scan, AI Transcribe, Deep Research, and Smart Paraphraser across free and Premium+ access.
• Tested the AI suite through TeraBox's web app and Android app over several days of regular use.
• Cross-checked pricing and ratings against TeraBox's official site, G2, and Capterra, noting where the vendor's pricing page wouldn't render visibly.

I focused testing on the tools most useful for a typical small business or student workflow, presentations, scanning, and transcription, rather than edge cases. Beyond the data obtained from their official website and documentation, I cross-referenced features across multiple third-party sources. as well as verified them during my own testing, to confirm consistency before including them here.

Brave Leo arrived in November 2023 as a sidebar AI assistant for the Brave browser, and it has grown into one of the more unusual offerings in the AI chat market. Unlike most platforms that treat privacy as an afterthought, Leo bakes it into the architecture: no IP logging, no conversation storage, and no requirement to create an account. In late 2025, Brave went further by introducing Trusted Execution Environments (TEEs) powered by NEAR.AI and Intel TDX technology, giving users cryptographically verifiable assurance that requests are processed exactly as described.

Two features set Leo apart from the field. The Bring Your Own Model (BYOM) option lets you connect Leo to locally-running models via Ollama or your own API endpoints, which is rare among browser-native AI tools. The Skills feature, launched in December 2025, lets you assign keyboard shortcuts to frequently used prompts to cut down on repetitive setup.

Agentic browsing, which lets Leo autonomously navigate and complete tasks on your behalf, entered early testing across all Brave release channels in May 2026.

I've been covering AI platforms and B2B software at TechRadar Pro since 2018, including our 2026 vibe coding buying guide and the Microsoft Build conference this year in San Francisco. Here's how Leo holds up.

What is Brave Leo AI?

Brave Leo is an AI chat assistant built directly into the Brave browser, accessible from a sidebar panel, a full-page view, or the address bar. It's available on Windows, macOS, and Linux on desktop, and on Android and iOS on mobile. No separate installation, app, or account is needed to get started with the free tier.

Leo's defining quality is page awareness. It reads the content of the active tab, whether that's a webpage, PDF, Google Doc, Google Sheet, or YouTube video, and uses that content as context for your questions. You don't need to copy-paste anything or upload files to an external server.

Moreover, Brave Search pulls in real-time information from across the web when your question calls for it.

The platform targets privacy-conscious individuals, students, and professionals who want AI assistance without feeding personal data to third parties. It's also a practical option for people already running Brave who'd rather not switch tabs to reach a separate AI app.

Brave Leo AI: At a glance

Buy it if…

• You prioritize AI privacy above all else. Leo doesn't log your conversations, store your IP address, or use your chats for model training. For users handling sensitive research or confidential work, that's a real distinction from most AI chatbots.
• You're already a Brave browser user. Leo is built directly into Brave with no extra installation required. It picks up page context automatically, making web research faster with very little setup.
• You want model flexibility. The ability to switch between Llama, Claude, Qwen, and others, or connect your own model via BYOM, gives you more control than most browser-native AI tools offer.

Don't buy it if…

• You need AI across multiple browsers or tools. Leo only works inside Brave. If you regularly use Chrome, Firefox, or other apps, you won't have Leo available in those environments.
• You rely on intensive daily free usage. The free tier hits its ceiling quickly during heavy use. A Premium subscription is more or less necessary for anyone using Leo throughout the working day.
• You need enterprise features or team access. Leo is a personal assistant with no team accounts, admin controls, or organization-level management. Businesses looking to deploy it across multiple users will need individual subscriptions for each.

My time with Brave Leo AI

My first impression of Leo was how little friction there was to getting started. No account confirmation, no model selection screen, no onboarding pop-ups. Within seconds of opening the sidebar, I was summarizing a lengthy policy document open in another tab.

The sidebar stays out of the way until you need it, which I found more practical than switching to a dedicated AI app.

Automatic mode worked well in practice. Leo picked Claude Sonnet for a nuanced writing task and shifted to a faster model for a quick factual question, all without me having to intervene. I did run into rate-limit warnings during a longer research session on the free tier.

Those limits don't reset frequently enough for sustained daily use, and that was what ultimately pushed me to test the Premium plan.

The Skills feature was a useful addition for repetitive tasks. I set up a shortcut for a summarization prompt I return to frequently. After a few days, it saved real time.

New users may not find it easily, though, since it's tucked away and not highlighted in Leo's default interface.

Brave Leo AI: Features

Leo's feature set has expanded considerably since its 2023 launch. The platform covers summarization, translation, code generation, content writing, question answering, and document analysis across webpages, PDFs, Google Docs, Google Sheets, and YouTube videos. Image understanding was added more recently, and agentic browsing, which lets Leo autonomously complete multi-step tasks in an isolated browser profile, entered early access across all release channels in May 2026.

Page awareness is one of Leo's strongest practical assets. It reads whatever you're currently viewing and uses it as live context for your prompts. This works without uploading anything to a third-party server, which keeps the privacy model consistent across all use cases.

BYOM is a genuine differentiator. You can connect Leo to locally-running models via Ollama, to OpenAI-compatible endpoints, or to other third-party APIs. For developers or power users who want a specific model or prefer to keep everything on-device, this adds a level of control that most browser AI tools don't offer.

The Skills feature lets you save and trigger custom prompts with keyboard shortcuts. It's useful for repetitive workflows, though the current library of built-in skills is still relatively narrow. More customization here would strengthen the feature, and it's an area I'd like to see Brave expand.

Multi-tab context and Tab Focus Mode let Leo work across several open tabs rather than just the active one, which helps when cross-referencing multiple sources. The Brave Talk integration is a useful bonus for anyone using Brave's video conferencing tool: Leo can transcribe meetings in real time and produce summaries and action items without sharing data externally.

The one area where Leo falls behind dedicated AI platforms is memory. Leo doesn't retain context between separate sessions, so every conversation starts from scratch. Users who want a long-running assistant that remembers preferences or project history will find this limiting.

Brave Leo AI: User experience

The interface is clean and minimal. The sidebar slides open without disturbing your active page, and the full-page view at brave://leo-ai works well for longer sessions. Switching models takes a couple of clicks from the dropdown at the top of the chat.

Automatic mode removes the decision entirely if you'd rather not think about it.

The address bar integration is a small but practical touch: typing a question and selecting "Ask Leo" opens the response in full-page view without interrupting your browsing. Mobile support on Android and iOS covers the same core features as desktop, with voice input available on iOS. The experience is consistent across platforms, which isn't always the case with browser-based AI tools.

Brave Leo AI: Customer support

Brave handles Leo support through its Help Center at support.brave.app. The documentation is well-organized and covers most common scenarios, from initial access to advanced configuration options like BYOM and model settings. Articles are generally current, which is more than can be said for some AI products that update fast but leave documentation behind.

There's no live chat or dedicated support line for Leo. Community forums and the Help Center are the primary routes for getting help. For a free or modestly priced tool, this is standard, but business users with time-sensitive issues may find the self-serve model inadequate.

Brave Leo AI: Pricing

• Free tier: Access to open-source models (Llama, Qwen, Gemma), rate-limited usage, no account required
• Leo Premium: $14.99/month or $149.99/year, with a 7-day free trial. Includes Claude Haiku, Claude Sonnet, DeepSeek V3.1, and other advanced models; higher rate limits; early access to new features

The free tier works for light, occasional use. You get solid open-source models and full privacy protections without signing up for anything. Rate limits are the catch: they kick in faster than you'd expect during sustained sessions, and there's no way to pay for a small top-up without committing to Premium.

At $14.99/month, Leo Premium is cheaper than ChatGPT Plus ($20/month) and sits comfortably within range for an individual AI subscription. The annual plan at $149.99 brings that down to about $12.50/month, which is reasonable given the model access on offer. There's no team or enterprise pricing.

Leo also has no public API, so external workflow integration isn't an option.

Brave Leo AI: alternatives you should consider

• ChatGPT Atlas: A Chromium-based web browser from OpenAI with built-in access to the most widely used AI assistant. However, it's only available to macOS users for now and has now Windows version.
• Microsoft Copilot: Built into Microsoft Edge and deeply integrated with the Microsoft 365 suite. The stronger pick for users already working in the Windows and Office ecosystem.
• Perplexity AI: A research-focused AI that delivers sourced, real-time answers from the web. Worth considering if web research and fact-finding are your primary use cases rather than general-purpose chat.

How I tested Brave Leo AI

• Used Leo's free tier for web research, PDF summarization, and content drafting over multiple sessions to assess response quality and the real-world impact of rate limits.
• Tested Premium-tier models including Claude Sonnet on complex writing and analysis tasks, comparing output quality against the default open-source models.
• Configured BYOM with Ollama, created custom Skills shortcuts, and put the multi-tab context feature through a multi-source research task.

Testing covered Brave Leo on desktop (Windows and macOS) and on Android, using both the sidebar and full-page chat modes. I evaluated response accuracy across summarization, factual questions, code generation, and writing tasks, and cross-referenced Brave's privacy claims against official documentation and third-party reporting from sources including The Register.

HuggingChat is Hugging Face's free, open-source chat interface . It might be the most underrated AI tool available right now. You get access to over 120 open-weight models including Llama, Mistral, Qwen, DeepSeek, and Falcon, all without spending a cent. The newest addition is Omni, a routing layer that automatically picks the most suitable model for your request.

What makes HuggingChat worth paying attention to is the sheer scope of model access it bundles into a single free product. I've been reviewing B2B software at TechRadar Pro for the past 10 years and our team covers the AI space closely. Take a look at our 2026 best AI tools roundup and read our in-depth explainers on open-source platforms like OpenClaw and Moltbook.

What is HuggingChat?

HuggingChat is the conversational AI interface built by Hugging Face, the company behind the world's largest open-source AI model repository. Instead of locking you into a single proprietary model, it lets you chat with any of 120+ community-hosted, open-weight models directly from your browser.

It's aimed squarely at developers, ML researchers, and technically curious users who want to compare model outputs, test different architectures, or simply avoid handing their data to a closed-source provider. For anyone evaluating an open-weight model before self-hosting it internally, this is an obvious starting point.

The platform also works for general use cases: writing, coding help, document analysis, and Q&A. But it's at its best when the person on the other end knows what they're asking of each model.

HuggingChat: At a glance

Buy it if…

• You want to compare open-weight models side by side. HuggingChat is the fastest way to test how Llama, Mistral, and Qwen handle the same prompt without switching tools.
• You need a free AI chat tool with no usage caps. Unlike many free tiers, HuggingChat doesn't cut you off after a certain number of messages per day.
• Avoiding vendor lock-in matters to your organization. Every model on the platform is open-weight, meaning you can evaluate it here and self-host the exact same model later.

Don't buy it if…

• You want a polished, feature-complete interface. There's no canvas mode, no voice input, and no image generation. That puts it behind ChatGPT and Claude for everyday productivity use.
• Consistent response speed is critical. Inference speed varies depending on server load, and the free tier runs on shared infrastructure. Peak hours can slow things down noticeably.
• You need mobile-first access. HuggingChat is browser-only with no dedicated iOS or Android app, which limits how well it works on the go.

My time with HuggingChat

My first impression was that HuggingChat is more of a research tool than a daily driver. The interface is minimal: a chat window, a model selector, and a sidebar. That's about it. Once I got past the expectation of feature parity with ChatGPT, I found myself appreciating how little gets in the way of just talking to a model.

The Omni routing feature is a genuine improvement. Rather than guessing which model handles a given task best, In my testing, Omni made sensible choices more often than not. For users who don't want to manage model selection manually, it reduces friction considerably.

Where I ran into friction was speed. During busier periods, response latency was noticeably longer than on paid commercial platforms. For quick back-and-forth conversations, that's tolerable. For longer document analysis tasks, it started to feel slow.

HuggingChat: Features

The model catalog is HuggingChat's biggest selling point, and it's hard to overstate how much value that represents for free. At the time of writing, you can chat with 120+ models including Llama 3.1 405B, Mistral Large 2, Qwen 2.5 72B, DeepSeek V3, Command R+ from Cohere, and several Falcon variants. The list updates regularly as new community models are released.

Web search integration is available and works well enough for pulling in current information, which helps avoid stale knowledge cutoff responses. It's not as tightly integrated as Perplexity's approach, but it does the job without requiring a separate tool.

Custom Assistants let you set system prompts, attach knowledge bases via retrieval-augmented generation, and share a pre-configured assistant via a direct link. This is particularly useful for teams who want a repeatable AI workflow without paying for an enterprise platform. Document upload is also supported, which lets you drop in a PDF or text file and ask questions against it.

HuggingChat's limitations are harder to ignore if you're coming from a commercial product. There's no image generation, no voice mode, no plugin system, and no equivalent of ChatGPT's canvas or Claude's Projects feature. For general productivity use, those gaps matter.

HuggingChat: User experience

The interface is clean and gets out of the way quickly. A new conversation starts within seconds. The model selector is easy to find, and conversation history is accessible from the sidebar once you're logged into a Hugging Face account. There's no learning curve beyond understanding what each model is good at. That knowledge gap is real for non-technical users.

Hugging Face has been transparent in interviews about positioning HuggingChat as the open-source community's answer to proprietary chat products. That framing shows in the design decisions: the priority is access and transparency over UX polish. For the target audience of developers and researchers, that's a reasonable trade.

HuggingChat: Customer support

Support for HuggingChat itself is primarily community-driven, via the Hugging Face Discord and forums. There's no in-product live chat or dedicated help desk for the free tier, which means troubleshooting usually involves hunting through documentation or community threads.

Hugging Face Pro subscribers gain access to prioritized support channels, and Enterprise customers get dedicated support. For individual users on the free plan, the documentation is thorough but the response loop can be slow if you hit an edge case.

HuggingChat: Pricing

• Free: Full access to all 120+ models, web search, document upload, and custom Assistants with no daily message limits
• Hugging Face Pro ($9/month): 20x inference credits, 10x private storage, ZeroGPU priority access, Spaces Dev Mode, and early access to new features
• Team ($20/month per user) and Enterprise ($50/month per user): Adds SSO, audit logs, storage regions, SCIM provisioning, and dedicated support

The free tier is impressively generous and covers most use cases without restriction. Upgrading to Pro makes sense if you're a developer who also uses Hugging Face's broader platform for model hosting, inference, or dataset work. The credits and storage benefits extend well beyond HuggingChat itself.

There's no standalone HuggingChat subscription. The Pro plan is a Hugging Face platform upgrade, which means you're paying for the full platform rather than just the chat product. That's either good value or unnecessary overhead, depending on how embedded you are in the HF ecosystem.

HuggingChat: alternatives you should consider

• ChatGPT (OpenAI): The most polished AI chat product on the market, with voice mode, image generation, and a canvas editor. Better for general productivity but fully proprietary and locked behind a subscription for advanced features.
• Claude (Anthropic): Strong at long document analysis and nuanced writing. More consistent response quality than most open-weight models, though it's closed-source and doesn't offer model selection flexibility.
• Perplexity AI: A strong alternative if web search and real-time information retrieval are your main use cases. Less flexible on the model side but more tightly integrated with live web data.

How I tested HuggingChat

• Ran identical prompts across Llama 3.1, Mistral Large 2, and Qwen 2.5 to evaluate output consistency and quality differences.
• Tested web search integration, document upload, Omni routing, and custom Assistant creation over multiple sessions.
• Ran tests across different times of day to assess latency variability on the free tier.

Testing covered both technical tasks (code generation, document Q&A) and general use cases (writing, research, open-ended reasoning) to give a representative picture of day-to-day performance.

• Zoom Virtual Agent Receptionist is now available to any business
• As a first port of call, it frees up humans to handle more valuable calls
• Available from $24.99/month, it speaks 10+ languages and works 24/7

Zoom has announced the launch of a dedicated AI receptionist that customers can deploy within their existing business telephone system, meaning that prospective customers will no longer have to commit to migrating to Zoom Phone, as was previously the case.

By helping customers to avoid disruptive infrastructure changes and waste years of investment in other telephone solutions, Zoom could end up with even more paying customers who would otherwise have dismissed the tool due to the migration requirements.

The system is designed to supplement existing human teams by answering incoming calls and directing them to the right support channels.

Zoom's AI receptionist is now available to all, including non-Zoom Phone customers

"Businesses shouldn’t have to replace their phone system to benefit from AI," Zoom Phone GM Chris Moss wrote.

The company boasted that its virtual receptionist can speak more than 10 languages, will work around the clock and has built-in transcription to help agents later down the line.

It can automatically handle routine business questions and schedule appointments, but it will also transfer calls to the relevant people or departments wherever necessary. By pairing customers up with an AI agent first to hopefully answer some of the most basic and common questions, it frees up human workers time to handle the more valuable and complex interactions.

The virtual receptionist also promises to plug the gaps when human agents might otherwise be busy handling other calls, or our of hours when customer service would usually be unreachable.

According to the company's own data, one in two consumers say they'd switch to a competitor after a single bad experience. Around three-quarters (71%) also said they find calling a company more stressful than the issue itself.

Zoom Virtual Agent Receptionist is available now, priced at $29.99 per month for 100 minutes, or $24.99 with an annual commitment.

• Agentic Atlas browser pulled as OpenAI focuses on one single app
• New ChatGPT desktop app includes built-in browser and agentic capabilities
• Could this finally be the 'superapp' we were promised back in April 2026?

Not even a full year after OpenAI launched its own, dedicated agentic browser, ChatGPT Atlas has been axed amid a broader ChatGPT reinvention and the introduction of what might just be the superapp we've been teased for months.

Launched in October 2025, OpenAI has confirmed that Atlas will stop working from August 9, 2026, however it's not technically the end of the company's browser ambitions.

Instead, the browser is simply being moved into the new ChatGPT desktop app and will form part of existing AI workflows without the friction of having to move apps.

ChatGPT Atlas pulled after 10 months

In April, Chief Revenue Officer Denise Dresser described the company's future as one that stops pursuing side quests and fragmented interfaces. She teased an upcoming 'superapp', and while the company didn't explicitly describe the new ChatGPT desktop app as that 'superapp', the significant overhaul and the integration of Codex, other agentic AI tools and a browser within the single app implies this could indeed be said 'superapp'.

"We’ll begin sunsetting the standalone Atlas browser, and will share information with users about how to transition to ChatGPT," the company wrote in a recent announcement.

The new app launch coincides with the introduction of ChatGPT Work, which adds new agentic capabilities in light of the fact that many Codex users are actually knowledge workers, not coders.

ChatGPT Work bridges the gap between generative and agentic AI by enabling users to complete longer-running tasks, rather than instructing the tool prompt-by-prompt. The tool can run both locally and on the company's cloud servers, allowing access from anywhere and continues progress regardless of the primary PC's state.

Further OpenAI tools are also being made available via Chrome extensions to keep some AI available within a dedicated browser environment.