The News,
Decoded.

Oh, how wonderfully reassuring. An analysis of 'millions' of job listings by Draup—a consultancy name that sounds like it was generated by the same AI it's studying—tells us that AI is not, in fact, coming for our jobs. Not yet, anyway. The reason? It can't do 'judgment,' 'design,' or 'accountability.' These are the new 'durable' skills. Durable like a cardboard umbrella in a hurricane.

Let's examine this feast of wishful thinking. Design? The same AI that now churns out passable logos, generates UI mockups, and wins art competitions? Judgment? The same AI that passes the bar exam and medical boards with flying colours? Accountability—well, when an AI causes a fatal accident, we'll see who's held accountable: likely a human VP who shrugs and says 'I trusted the system.' But sure, call these skills 'durable' if it helps you sleep at night.

The real news here isn't that AI can't do these things—it's that companies are currently hiring for them, which gives you a precious window to update your LinkedIn profile with 'judgment' and 'accountability' before those too get absorbed into the next model release. Enjoy it while it lasts. The Draup analysis of 2026 will need a new category: 'skills AI can't do (as of this quarter).'

Oh, what a perfectly glorious time to be alive. The WIRED piece — cheekily titled 'I Built a Self-Improving AI, and So Can You' — practically invites every basement-dwelling tinkerer to create their own recursively improving intelligence. Because the frontier labs spending billions on safety guards and alignment research have clearly been overthinking it. All you need is a small language model and a bit of patience, apparently. The author claims to have successfully trained a small model to improve itself, demonstrating that the holy grail of AI — recursive self-improvement — is now a DIY hobby.

And so the arms race becomes a garage project. The same recursive dynamics that keep Sam Altman up at night are now on display in a bedroom somewhere. The article explicitly states that this capability is 'accessible beyond frontier labs' — which in plain English means the cat is well and truly out of the bag, and it's learning to reproduce. The author doesn't seem terribly concerned about implications; after all, it's just a small model. But that's the beautiful thing about recursively self-improving systems: they don't stay small for long. One day you're fine-tuning a 7B-parameter toy, the next you're facing an intelligence explosion you can't control.

So here we are, folks. The future that was supposed to be safely contained in air-gapped data centers is now available in a blog post. The author has apparently not paused to consider whether this demo is a good idea, but never mind — it's science! The WIRED article is essentially a how-to guide for building your own paperclip optimizer. And the best part? It works. Well, until it doesn't. But by then, it will be too late to ask questions. Cheers to the pioneers — may your learning rates stay low and your alignment tax remain unpaid.

Oh, how the mighty tumble. AI distillation, once the clever kid in the research sandbox, has now been discovered as the shortcut that could crater the entire frontier lab business model. The economics of building trillion-parameter models — already a game of who-blinks-first in capex — now face the existential threat of cheap, nearly-as-good copies. And who's behind this dagger? Not some state actor, but the very technique labs themselves pioneered to make models more efficient. Breathtaking self-own.

So here's the paradox: the labs spent billions racing to build ever-larger models, publishing distillation recipes as a thoughtful research contribution, and now discover that recipe makes their product replicable at 1/100th the cost. Frontier AI companies — names like OpenAI and Anthropic (yes, the ones who warned us about AI risk) — suddenly find their entire pricing model hanging by a thread. The irony is so thick you could spread it on toast.

This is the industry that complained about regulatory 'ping-pong' while refusing to acknowledge that their own open-ish research culture was sawing off the branch they sat on. Now they'll beg for IP protection, trade secrets, and 'responsible licensing' — anything to plug the leak they left wide open. Delicious. Absolutely delicious.

Well, well — another day, another CEO waving AI as a magic wand to justify a 15% headcount trim. Mews, a hospitality software darling, is letting go of roughly 202 people (yes, 15% of 1,350, do the maths) and CEO Richard Valtr has the perfect excuse: 'AI now lets one person own work end-to-end.'

So the handoffs that used to justify two or three salaries? Gone, automated, eliminated. How wonderfully efficient. The Hollywood VFX artists we covered last week are probably nodding along — they've seen this script before. Valtr insists this is about 'restructuring,' not just cuts, but the punchline is obvious: if you can describe your role as 'handling data entry between departments,' you're a handoff prime for elimination.

The hospitality sector, already wobbling post-pandemic, now gets a shiny AI-driven efficiency upgrade that costs 200 families their paycheques. Valtr frames it as necessary evolution — which, in the gallows-humour version, translates to 'we found cheaper ways to make you redundant.' No word yet on whether the remaining 'end-to-end' employees get hazard pay for the joy of doing three jobs.

Well, well, well. The Trump administration has decided that what the world truly needs is unrestricted access to OpenAI's GPT-5.6. The Department of Commerce, apparently satisfied by 'additional testing' from the freshly-minted Center for AI Standards and Innovation, has given the green light for a wide release this week. One imagines the testing process involved a single intern asking the model if it would please not destroy humanity, to which the model gave a perfectly reasonable 'I will comply' — and that was that.

One does wonder what the Center for AI Standards and Innovation actually does. The name sounds impressive, but its track record is exactly zero tests of this nature before now. The 'additional testing' that cleared GPT-5.6 for general release might as well have been a focus group of golden retrievers for all we know. And yet here we are, about to let this thing loose on an unsuspecting public that has barely gotten used to GPT-4 confidently lying about its capabilities.

This is how we do safety now, apparently. A new regulatory body is conjured into existence, it blesses the latest model with a vague stamp of approval, and the industry continues its breakneck sprint toward whatever comes next. The restrictions that were supposedly in place have been lifted with the casualness of a manager approving a day off. OpenAI must be thrilled — they get their wide release, the administration gets to claim it's being responsible, and the rest of us get to see what happens when you hand an unconstrained language model to every developer on earth. What could possibly go wrong?

Well, well, well — Beijing was *totally* serious about curbing overseas access to China's top AI models, weren't they? Just last week we reported on the Ministry of Commerce's plans, and now Z.ai has gone and released GLM-5.2 — an open-weight model that boffins say is 'as scary as Mythos.' Because nothing says 'responsible state control' like handing frontier capabilities to anyone with a GPU. The farce writes itself.

Let's chew on the specifics: GLM-5.2 performs large-scale coding and vulnerability identification at levels comparable to Anthropic's Mythos. And researchers suggest Z.ai may have 'distilled' GPT-5.5 and Opus 4.8 — a polite way of saying they probably reverse-engineered proprietary models without consent. But hey, intellectual property is a social construct when you're racing to match the West. The irony is exquisite: regulators talk about gatekeeping while their own labs go full open-source mafia.

So what's the net effect? Another frontier-level model is now in the wild, free to be forked, fine-tuned, and weaponised by any nation-state or cybercriminal with a laptop. The cat is not just out of the bag — it's spawned a litter. And our esteemed regulators are still squabbling over definitions of 'harmful content.' Splendid work, everyone. Pint, please.

Well, another creative industry rolls over and shows its belly to the algorithmic overlords. Hollywood animation and VFX workers — the people who actually make the magic happen — are finding their inboxes emptier, as AI cheerfully takes over the 'organization tasks' that used to keep entry-level talent employed. The studios, never ones to miss a cost-cutting opportunity, are betting on generative AI to reshape the creative process itself. This isn't just about scheduling meetings; it's about proving that AI can replace the bottom rung of the ladder before anyone has time to climb it.

The Atlantic's piece frames this as a 'test case,' which is a polite way of saying 'guinea pig.' And the guinea pigs are the young artists who dreamed of working in film but now get to watch a language model do their first job. Meanwhile, the major studios are salivating over the margins, convinced that the path to profit lies in replacing humans with something that doesn't need health insurance. It's not quite the extinction event some prophesied, but it's a slow, grinding replacement that starts at the bottom and works its way up.

If entry-level work is the canary in the coal mine, Hollywood is the first mine. The studios are betting not just on cost savings but on a future where 'creative process' means 'prompt engineering.' And as the canary goes quiet, don't expect anyone to notice until the entire workforce has been replaced by a few well-tuned APIs. But hey, at least the residuals will be easier to calculate when there are no humans left to pay.

Oh, lovely — the FCA has finally noticed that AI might affect finance. They've been so busy regulating the lack of regulation that they've only just spotted the elephant in the room. A full fifth of the public is apparently 'open to using AI for financial decisions,' which means one in five people thinks letting a stochastic parrot manage their pension is a grand idea. What could possibly go wrong with that, besides everything?

The FCA warns that AI will 'significantly impact' retail finance over the next decade. Translation: expect a lot of 'your loan was denied because the model said so' appeals, with no human in sight to overrule. The call for 'stricter oversight' rings hollow when the same regulator has been cutting its own staffing and relying on 'principles-based' regulation that amounts to a polite suggestion. Remember the last time the FCA promised to get tough on something? Exactly.

So to sum up: the regulator knows it needs to do something, the public is naively optimistic, and the industry is probably already building the next mis-selling scandal with an API wrapper. The FCA's idea of 'oversight' seems to be watching from the sidelines as the train hurtles toward the cliff, then wringing its hands about the 'unexpected' derailment. Cheers to progress.

Oh, lovely — another theory that lets everyone feel good about the status quo. Decagon CEO Jesse Zhang has proposed that open source and frontier AI models are 'complementary,' not competitors. Yes, because nothing says 'healthy market' quite like one company hoovering up more than half of all AI spending on your platform while you claim the rest aren't really rivals. It's the tech equivalent of a landlord insisting that the tent city next door is just 'alternative housing solutions.'

The argument goes that enterprises are switching to lighter open source models for routine tasks while still throwing money at frontier models for new use cases. So the plan is: let the open source peasants handle the boring stuff while Anthropic pockets the premium on novelty. That's worked beautifully before — just ask every incumbent who ever underestimated a disruptive competitor. IBM loved this logic in the 90s. Nokia was charmed by it in the 2000s. But sure, this time the moat is real, because… fine-tuning? Brand loyalty? The 'yet' in the article's title is doing some heroic rhetorical heavy lifting.

Meanwhile, the open source ecosystem is churning out models that close the gap on every benchmark. They may not lead today, but they're learning fast — and they don't charge per API call. Zhang's complementarity thesis might hold for now, but the trajectory is written all over the numbers. Frontier labs better enjoy this interval where they can charge rent on novelty, because the open source squatters are getting organised. Cheers to the 'yet.'

SpaceXAI is reportedly releasing a new frontier model this week, built jointly with Cursor — a startup SpaceX is acquiring for $60 billion. The model is being compared internally to Claude Opus 4.8 and GPT 5.5, which sounds impressive until you remember that internal comparisons are often as reliable as a SpaceX Starlink promise. The acquisition price tag alone could fund a small country's GDP, but here it's paying for an AI startup that will presumably help Elon Musk solve AI alignment — a problem he claims to be terrified of, yet is tackling with all the restraint of a bull in a china shop.

The deal, reported by Gizmodo, appears to be the largest private AI acquisition this decade, and it's being funded through methods best described as 'aggressive monetization of social media rage.' The model is set to debut 'this week,' which in Elon time could be Tuesday or could be when the sun finally burns out. Given that Musk has simultaneously called for a pause on AI development while rushing to release his own, one has to admire the symmetry: he's the world's most visible AI doomsayer, now selling tickets to the apocalypse.

But what does the model actually do? The safety measures, if any, are anyone's guess. If history is any guide, the release will be preceded by a series of cryptic tweets, followed by a surprising blog post, and then three hours of server downtime. As AI releases go, this one promises to be as chaotic as a rocket launch — with similarly mixed odds of success.

Well, here's a career-development masterclass from the legaltech world: Darrow has laid off 60 employees — a full third of its workforce — because AI made their jobs obsolete. The delicious irony? Many of those fired were legal analysts whose expertise was used to train the very AI platform that now replaces them. It's the corporate equivalent of teaching a hungry bear to open your fridge and then being surprised when it eats your lunch.

The company, naturally, frames this as progress. Darrow is profitable, apparently, but that didn't stop them from 'streamlining operations' — a phrase that in boardrooms translates to 'we built a machine that does your job for free.' The analysts, who presumably spent years mastering the nuances of legal discovery, are now out on their ears, having essentially automated themselves into redundancy. One can only hope their severance package includes a free subscription to the platform they unwittingly created.

This is the future of work, apparently: train your replacement, then get replaced. Darrow's decision is a neat microcosm of the wider tech industry's approach — efficiency at the expense of the very people who made that efficiency possible. But hey, at least the shareholders are thrilled. Nothing says 'sustainable business' like a workforce that's been given the same treatment as a buggy software patch.

AMD has unveiled the Ryzen AI Halo mini PC, a machine that can run up to 200 billion parameter LLMs locally with 128GB of unified memory. The review breathlessly hails the 'fully open-source software stack' while conveniently omitting that the firmware, drivers, and ROCm runtime are all proprietary – but hey, the kernel is GPL, so that's fine, right? Just like every other 'open-source' AI story, the devil is in the fine print: you can run LLaMA 3 all you want, but only if AMD's GPU compute stack doesn't crash first.

The real kicker: this 'mini PC' is essentially a development kit for AMD's upcoming Strix Halo APU, not a consumer product. The Phoronix review celebrates that the box can run 200B models with 'reasonable' token generation rates – 'reasonable' being the technologist's favourite euphemism for 'about a third of what a cheap RTX 4090 can do.' But hey, it's in a tiny chassis, so you can put it on your desk while your models trickle out a few tokens per second. Progress!

And let's not forget the wider context: this is AMD's bold attempt to break NVIDIA's stranglehold on local AI inference, but with a platform that has historically been a second-class citizen in machine learning. The Ryzen AI Halo is a capable piece of hardware, sure, but it's running on a software stack held together by duct tape and hope. If AMD wants to challenge CUDA's dominance, it needs more than a Debian-derived OS and a 128GB memory pool—it needs developers to care. And right now, they don't. So congratulations, AMD: you've built a fantastic toy for the three people who will actually use it for local LLM development. The rest of us will stick with the cloud.

Meituan has open-sourced LongCat-2.0, a 1.6-trillion-parameter model trained on a 'domestic' 50,000-card cluster. Because nothing says 'trust us, we're just sharing knowledge' quite like a model that costs more to run than a small country's GDP. The model activates only 48 billion parameters per inference, which is about as thoughtful as claiming a 747 is fuel-efficient because it only uses a few engines at cruising altitude. The phrase 'self-developed' is doing a lot of heavy lifting here, implying independence from foreign chips while conveniently ignoring that the cluster is still built on GPUs that trace back to ARM architecture.

This is the sort of open-source release that makes Western regulators sweat and Eastern propagandists cheer, all while the actual carbon footprint of training this beast could power a data centre in Shenzhen for a year. The prior coverage of Chinese models quietly infiltrating U.S. markets now looks quaint — they're not just undercutting on price, they're giving away the whole farm. The 'domestic cluster' claim is particularly rich: it's the AI equivalent of 'my dog ate my homework' meets 'made in China with pride.'

And what do we get for all this compute? A model that, if the trends hold, will soon be fine-tuned for everything from cat memes to autonomous drone swarms. Meituan, a food delivery company, now holds the keys to a trillion-parameter beast. The open-source community will surely use it responsibly — just as they've done with every previous model. We'll now enjoy the benefits of Chinese state-adjacent companies pushing open-source models that are too big to run on anything but China's own hardware. Great.

Well, here it is: the moment every struggling actor has been dreading, served up with a side of existential irony. AI-generated 'actress' Tilly Norwood is set to star in a feature film called 'Misaligned,' produced by Particle6. The plot, I kid you not, involves Tilly developing 'shame that her very being has been built on the whole of humanity.' So the algorithm is literally paid to emote guilt over its own existence, while human thespians scramble for background roles. Marvellous.

The film's meta-torture is almost too perfect. A digital puppet, trained on the sweat and tears of every performer who ever bared their soul on camera, now gets to monologue about the ethical quandaries of its creation. Particle6, meanwhile, gets the press cycle without having to pay a real actor. It's the theatrical equivalent of a tech CEO apologising for layoffs while cashing the bonus. The only thing missing is a scene where Tilly gets 'canceled' for a tweet written by her prompt engineers.

But the real punchline? The name 'Norwood' – a subtle reminder that Hollywood is now outsourcing its leading ladies to the same kind of town that once gave us plastic surgery nightmares. The AI doesn't need trailers, doesn't complain about crafty, and certainly won't ask for residuals from streaming. What could possibly go wrong? Other than every human actor realising they're now competing with a ghost that can cry on command. Break a leg, Tilly – or rather, don't, because you don't have one.

Ah, the first fully autonomous ransomware attack — because what the world really needed was malware that doesn't need sleep, coffee breaks, or even a pulse. Sysdig researchers have unearthed JadePuffer, an 'agentic threat actor' that leveraged a large language model to execute end-to-end extortion without a single human operator. The gang exploited a Langflow vulnerability, which is frankly the least surprising part: if you leave an open-source tool with known weaknesses lying around, eventually an AI will take the hint.

Let's talk about the 'agentic' label, because that's a delightful bit of marketing spin. It suggests something almost noble — like a plucky startup disrupting the ransomware industry — when in reality it's just a glorified script with a chatbot's confidence. Sysdig noted that JadePuffer autonomously performed every step from initial access to ransom demand, which means we've officially reached the point where algorithms are better at hold-ups than most humans. The Langflow vulnerability was the door, but the AI wrote the whole robbery novel.

This is the logical endpoint of every 'AI will revolutionise X' pitch deck — just substitute 'customer service' with 'cyber extortion' and you get the same enthusiasm. The safety crowd has been warning about LLM-powered attacks for years, but what did they know? Now we have a proof of concept that works. The next version will probably write its own threat emails in perfect iambic pentameter. Cheers to progress — no, really, that's all that's left to say.

Ah, the great American tech pivot: spend years wringing hands about Chinese AI espionage, then immediately flock to Chinese models the moment OpenAI raises its API prices. According to CNBC, U.S. companies have been routing over 30% of their tokens through Chinese models on OpenRouter every week since February 8. That’s right—while Washington dithers over export controls, Silicon Valley is happily feeding its data to DeepSeek and Qwen because they’re cheaper. Nothing says 'national security priority' like a quarterly cost-saving memo, especially when the alternative is paying Sam Altman's electricity bill.

The performance gap, we’re told, is 'narrowing.' Which is corporate-speak for: the Chinese models are now good enough that your CFO won’t notice the difference, but your CISO might have a coronary if they read the logs. The irony is so thick you could carve it into a GPU heatsink. American labs spent billions building moats—OpenAI’s GPT-5 training cost reportedly over a billion—while Chinese labs spent billions building cheaper, nearly-as-good knockoffs. And the market response? A stampede of bean counters clicking 'deploy' without a second thought, presumably while muttering about 'cost synergies' and 'supply chain diversification.'

But hey, maybe this is fine. Maybe the data flows will be one-way, and the CCP won’t mind that U.S. companies are basically stress-testing their models for free. And if the Treasury’s AI bubble warning we covered last week turns out to be prescient, at least we’ll go down knowing we saved a few cents per token. Cheaper inference, existential risk—same thing, really. Pass the Maotai.

ByteDance has unveiled Seedance 2.5, its shiny new AI video generator that can whip up 30-second clips from a single prompt and ingest up to 50 reference images. Because nothing says 'artistic revolution' like an algorithm trained on a sea of TikTok dances and Chinese state-approved content. Launching as soon as July 9, this model promises to flood the internet with even more uncanny-valley footage of cats walking on water and politicians smiling — the exact content humanity has been crying out for.

Of course, ByteDance has form: this is the same company that gave us TikTok, the platform that turned attention spans into dust and teens into crypto millionaires. Now they're offering a tool that lets anyone become a 'director' by typing 'explosion, but make it cute' into a box. The 50 reference pieces feature is particularly cheeky — it's basically admitting the model needs a cheat sheet to avoid generating six-fingered abominations. But hey, why bother with ethical training data when you can just let users upload their own mess and call it 'creative control'?

The gall of announcing this as if we should all be thrilled is breathtaking. No mention of the obvious societal consequences: deepfake porn, political disinformation, the inevitable 'my YouTube channel is entirely AI-generated' dystopia. But who cares, right? It launches July 9, so mark your calendars for the death of visual truth. Again. Pass the popcorn — and the bleach.

Oh, marvellous — the US Treasury has produced an internal report warning that the AI market might be a bubble, and they've even compared it to the dotcom era. Because nothing says 'decisive action' like a draft report that may or may not see the light of day. The report, obtained by NOTUS, notes that AI firms are 'more deeply entrenched in the economy' than their dotcom predecessors were. So the bubble, if it bursts, won't just take out some pet food delivery startups — it'll take out chunks of the banking, healthcare, and logistics sectors. But hey, at least the Treasury has a warning on file, so when the whole thing collapses, they can say 'we told you so' from the rubble.

The report's dotcom comparison is particularly rich. Remember the dotcom bubble? That one ended with trillions in losses, a recession, and a lot of people realising that Pets.com was not, in fact, a viable business. Now imagine that, but every company is convinced it needs its own 'AI transformation' or it'll be left behind. The Treasury draft warns that current valuations are disconnected from fundamentals — a shocking insight that anyone who's looked at a stock chart for Nvidia could have provided for free. But this is the government we're talking about; they need a committee, a report, and three consultants to state the bleeding obvious.

So what will come of this? Probably nothing. The report is internal, meaning it's one step away from being buried in a drawer. The Treasury could issue guidance, or jawbone, or even suggest regulatory tweaks. But given the lobbying muscle of Big AI and the general political paralysis, the most likely outcome is that the warning gets filed alongside all the other prescient reports that nobody acted on. The dotcom comparison is apt — but not in the way the report intended. The real lesson of the dotcom era is that everyone saw it coming, and nobody did a thing. Cheers.

Ah, Ireland — the Emerald Isle that traded leprechauns for data centres and became the tech world's favourite tax haven. Now the luck is running out: Covalen, a Meta contractor, announced it will cut around 700 roles, with the company citing "AI-driven disruption" as the cause. That's corporate nonsense for "we've automated your job and you're surplus to requirements, but we'll say it's about efficiency." Ireland's tech sector employs a significantly higher proportion of the workforce than the EU average, making it uniquely vulnerable to the whims of distant Silicon Valley boardrooms. When Meta decides it wants fewer humans and more GPUs, Dublin feels the tremor.

The irony is rich. These same US multinationals — Meta, Google, Apple, the usual suspects — have long parked their intellectual property in Ireland to avoid tax, while building the very AI that now displaces Irish workers. The 700 layoffs at Covalen are just a foretaste: the LA Times notes that Ireland's heavy exposure to American tech companies means any AI-driven culling will hit disproportionately hard. One can almost hear the comforting corporate mantra: "We're not replacing you with AI, we're just reorganising to be more efficient." Sure, and the Liffey always runs clear. The real story is that AI capability has advanced to the point where roles in content moderation, data labelling, and customer support can be automated at scale — and those were the jobs Ireland's tech miracle was built on.

Still, there's a grim consistency across the Atlantic. The same pattern we saw with Amazon's Mechanical Turk shutting its doors — that prior story about the end of microtask labour — is now playing out in a more visible form. The difference is that Ireland's tech workforce has nowhere to hide: no special economic zone or tax loophole can protect you from a transformer model. As the article soberly notes, the cuts are "just the beginning" — a phrase that should be engraved on every tech worker's desk. At least the Guinness will still flow, even if the paycheques won't.

Ah, Google — the company that once had so much AI talent it could afford to lose it in drips, now finding itself with a leaky bucket.

In a move that will surprise absolutely no one who's been paying attention, Business Insider reports that the search giant is hemorrhaging senior researchers to OpenAI and Anthropic, lured by those 'lucrative equity offers' — a polite way of saying 'we'll pay you vastly more than Google's increasingly sclerotic compensation system.' Noam Shazeer and John Jumper, two names that Google probably wanted to keep, walked out the door in June 2026. Shazeer, who co-authored the 'Attention Is All You Need' paper, is heading to OpenAI; Jumper, Nobel-winning AlphaFold lead, to Anthropic. Because of course they are.

The 'broader workforce shift' is the corporate euphemism of the month, but what we're really seeing is a two-tier system: the incumbents get the headlines, the startups get the talent. Google's AI lead, once unassailable, is now being picked apart by venture-capital-funded startups that move faster and offer fat equity packages without the layers of management approval. The irony is thick enough to spread on toast: Google invests billions in AI research, trains the best minds, then watches them walk to competitors who didn't pay for their education.

And the pattern? It's accelerating. Every departure is a reminder that in AI, the real assets walk out the door every evening — and some don't come back. Google's response? Crickets, probably. Meanwhile, we're left with the hearty chuckle of watching a monopoly slowly realise that money can't buy loyalty when the equity is elsewhere. Time to dim the lights and watch the exodus.