Future of Work Lab

The Human Signal.

We read where AI is taking work — and keep people at the center of it. Search the full library, or browse by day.

Wednesday, 22 July 2026

Demographics & Labor AI at Work Organizational Change

Companies are cutting jobs to bet on AI. The ones winning are hiring instead.

In the first half of 2026 about 82% of the world's tech layoffs came from US companies, many cutting staff on the promise that AI would fill the gap, according to a TradingPlatforms report. But the bet keeps misfiring: Ford quietly rehired more than 350 veteran engineers after its AI quality systems could not hold up on their own. The firms actually gaining from AI, PwC and Ramp find, are the ones using it to make their people stronger, and the heaviest AI investors grew their headcount by about 10%.

For youSort your own work into two piles, what you would happily hand to AI, and the harder judgment a company could not quickly rehire. Get visibly better at the second pile. That is the difference between a role that gets cut and one that gets called back.

Source: The Deep View ↗

Open Models AI Governance & Regulation Geopolitics & Security

Whether cheap, capable AI stays in your reach is being decided right now

The most powerful AI you can download and run yourself, for free, now comes mostly from Chinese labs like Moonshot, maker of the new Kimi system. US officials briefly weighed making those free AIs harder for Americans to use, then backed off, Axios reported, and the government's own review urged testing them rather than banning them. This matters to you because free, downloadable AI is the floor that keeps every provider's prices honest and lets ordinary people run private work without paying a meter for every task.

For youDo not tie your work to a single AI provider. Learn to move the same task between two or three tools, so no ban, price rise or outage gets to decide for you what you can still do.

Source: Axios ↗

Organizational Change AI at Work Skills & Reskilling

Managing more people used to mean more power. Now it is controlling your own work.

For years your level, your pay and your standing came from how many people reported to you. As AI takes over the scheduling, reporting and hand-offs that middle managers used to run, companies are keeping fewer managers over bigger teams, and Gartner expects one in five of them to cut more than half of middle-manager roles. So the title on the door now says less about real power than the quiet authority to set your own terms: what you take on, how fast, how you are judged, and when you walk away.

For youYour title is about to say less about your power than it used to. What still counts is control, the workload you set, the deadlines you can move, who sees your results. Write down where you already have that, and defend it harder than the next promotion.

Source: Charter ↗

Human Agency AI at Work

Stop asking AI for answers. Start letting it question you.

Most of us treat AI like a very fast encyclopedia: we ask, it answers, we move on. The AI researcher Francesca Rossi argues the more valuable move is the opposite, an AI that asks you the hard questions the way a good mentor or a Socratic teacher would. Instead of handing you a tidy conclusion, it asks which values are in tension, who is affected, and what would change your mind, so you leave the conversation thinking more clearly rather than less.

For youAn answer ends your thinking; a good question extends it. Point your AI at a real decision and tell it to ask you five questions before it gives any answer. The questions are usually the part you were missing.

Source: Francesca Rossi ↗

Frontier Models Skills & Reskilling

AI helped crack an 87-year-old math problem, in the hands of a world-class expert

A problem that had resisted mathematicians since 1939, the Jacobian conjecture, just fell in three variables. The mathematician Levent Alpöge put it to Claude Fable 5 and got back a short polynomial counterexample that other mathematicians have since checked; the two-variable case is still open. Notice who was needed: not only the AI, but an expert who knew which question to ask and could tell a real answer from a convincing fake.

For youGo deep in one thing rather than dabbling in ten. AI can produce endless answers; the lasting skill is telling a right one from a convincing fake. That kind of judgment only comes from real depth.

Source: Levent Alpöge ↗

Monday, 20 July 2026

Open Models Skills & Reskilling

Betting your career on access to the cleverest AI is the losing bet

A new free AI from China called Kimi K3 now matches the best paid systems from the United States, and it will not be the last. When the cleverest AI is cheap and everywhere, being the person who can get hold of it stops being worth much. What holds its value is the thing sitting on top that a cheap AI cannot copy: your customers, your own record of how the work really gets done, and your judgment about what is worth doing at all.

For youThe tool is not your edge. Your knowledge is. Spend one evening writing down how you really do your work, the judgment calls no manual covers. Then talk to your best customer until you know their problem inside out. That is what still earns money when the clever part costs pennies.

Source: Exponential View ↗

Demographics & Labor AI at Work

Work is getting more productive, but the gains are not reaching your pay

American workers are producing more per hour than they have in over twenty years, yet their pay, after inflation, is rising at half that speed, and the share of the economy that goes to workers just hit a record low. So far this is happening before AI has even shown its full effect, which tells you the gains from doing more are flowing to owners, not to the people doing the work. The real question of the next few years is who keeps the value once AI makes each person far more productive.

For youDid you produce more last year than the year before, and did your pay move with it? Where those two drift apart, start naming the value you create out loud at review time. Do it before the story hardens that the machine did the work and the savings belong to the company.

Source: The New York Times ↗

Skills & Reskilling AI at Work

AI can polish your job application, and that is starting to backfire

Job seekers now lean on AI to write their CVs, cover letters and even test answers, and more than half already do. The trouble is that when every application looks equally strong, it stops telling an employer anything, and managers say they are making more hires they regret. Hiring is shifting towards proving you can actually do the work, on the spot, rather than how good your paperwork looks.

For youLet AI tidy your CV if you like. Every other applicant is doing the same. What sets you apart now is one skill you can prove live, in the room, with no help at all. Pick that skill and practise it until an interviewer can see it for themselves.

Source: Bloomberg ↗

AI at Work Creative Work

Making a film just got twice as fast, and the crew is smaller for it

Netflix told investors that around 300 of its shows and films already used AI to build crowd scenes, battles and backgrounds that a crew used to make by hand. On one show it produced 17 minutes of finished footage twice as fast and for half the cost. The parts of a production that were pure grind are being handed to AI, which leaves the human job concentrated in taste and story rather than the long hours of manual work.

For youDraw a line through your own work. On one side the craft you would defend, on the other the grind you would gladly hand over. The grind is what AI takes first, so your standing shifts onto taste, the judgment that decides whether the fast, cheap version is any good. This week, practise judging quality in the half you are keeping.

Source: The Verge ↗

AI at Work Organizational Change

Say you used AI and your boss may quietly credit the machine, not you

There is a strange new penalty at work: when you admit you used AI, people often assume the AI did the hard part and you did little. Researchers found this can quietly cost workers on reviews, promotions and raises, even when the finished work is strong. The lesson is not to hide the tool, but to be clear about the judgment that was yours, what you decided, changed and threw out.

For youDo not hide the tool. Show the thinking. Whenever you share AI-assisted work, add one plain line about what you decided, changed or threw out. Credit does not follow whoever used the tool. It follows whoever can show the judgment behind the result.

Source: arXiv ↗

Saturday, 18 July 2026

AI Governance & Regulation Geopolitics & Security

AI's builders say the rest of us should decide what it is for

Demis Hassabis, the head of Google DeepMind and one of the people racing to build far more powerful AI, says its hardest questions are not technical, and should not be left to people like him to settle alone. What he actually proposes is narrower: a watchdog that vets the most powerful AI before release, funded mostly by the industry but overseen publicly, much like the body that polices stockbrokers, starting in the United States in the hope others follow. The deeper questions he leaves open, how people will earn and find meaning and whose values the technology carries, are ones he says all of society must work out together, in a short window before this AI arrives.

For youHis plan stops short of one thing worth pushing for, that every country, not only the US labs and the US government, has its own body where ordinary communities help decide how AI is used, so when you hear AI is being handled, ask whether anyone who shares your values is actually in the room.

Source: Demis Hassabis ↗

Skills & Reskilling AI at Work

The narrow job is splitting open, and the broad one pays more

It used to pay to specialize in one slice of a job, like sourcing candidates or writing ad copy. Now the AI handles the routine parts of those slices, so one person can run the whole cycle from start to finish. The people gaining are the ones who widen out and connect the pieces, not the ones who guard a single narrow task.

For youList the tasks a teammate would have handled for you three years ago, then pick one and learn to do it yourself with AI this month, because the roles next to yours are the ones quietly merging into your own.

Source: The Batch (Andrew Ng) ↗

AI at Work AI Agents

Your AI is moving out of the chat box and onto your computer

For a couple of years the assistant lived in a chat window and you copied its answers out by hand. OpenAI's new desktop app changes that: it opens your files, controls other apps and the web, and can run on a daily schedule until a job is finished. The useful part is knowing which mode to point it at, everyday office work in one and building software in the other, so you hand it the right kind of task.

For youBefore you let it run, set it to ask permission for each step instead of full access, because the moment it can touch your files and click around your screen, the skill that protects you is deciding what to hand over and checking what it did.

Source: Leon Van Zyl ↗

AI at Work Organizational Change

Across very different jobs, AI is moving from helper to doer

In medicine, banking, film, real estate, and software, the same shift is showing up: AI is starting to run core parts of the work, not just assist with them. McKinsey pulled these into a short set of explainers, one per industry, so you can see what is changing in yours rather than guess. Seeing the pattern early is how you decide what to double down on and what to let go.

For youWatch the explainer for your own field this week and write down one task in it that is moving from assist to done, because that single task is where your role either grows around the AI or gets absorbed by it.

Source: McKinsey ↗

AI Coding Tools AI at Work

You can now build a real website inside Canva, no extra subscription needed

Making even a simple website used to mean paying for a separate website builder, or hiring someone who could. Canva has added a feature that turns a plain description of what you want into a real, working site you can then edit like any other design, inside the same tool many people already use for slides and social posts. The quiet part is the saving: the thing you would have paid a separate subscription for now sits inside one you may already have.

For youBefore you renew or buy another online tool this month, check whether something you already pay for has quietly added the same feature, because more of these tools are folding in jobs that used to need their own subscription.

Source: Canva ↗

Friday, 17 July 2026

AI Agents Organizational Change

The price per question is the wrong way to judge what AI costs you

Most people still size up an AI tool by its sticker price, what it charges per question or per page. McKinsey argues that number has stopped meaning much, because newer systems fire off many hidden steps behind one request, so a cheap looking tool can quietly run up a big bill and a pricey one can pay for itself. The question worth asking is no longer what a request costs, but what the finished result is worth to you.

For youPick one AI tool you use, stop watching its per use price for a week, and instead track the hours or the rework it actually saves you, then decide if it earns its place on value rather than on the number on the invoice.

Source: McKinsey ↗

Open Models Frontier Models

The most powerful AI is becoming free to run and shape yourself

Until now, the best AI lived behind one company's paywall and you rented it one question at a time. This week two of the strongest open systems yet, from Thinking Machines and from Moonshot, were released for anyone to download, run, and adapt for free. When the raw intelligence gets this cheap and common, the value shifts to what a copy of it cannot give you: your own data, your customers, and your judgment about what is worth doing.

For youPicture where this goes, every person and every company running a model that costs nothing, sits on their own machine, and knows only their work, their customers, and their way of doing things. The intelligence in that picture is free. The record of how you actually work is not, and it is the only reason your copy would be worth more than anyone else's, so start keeping it now.

Source: Thinking Machines ↗

AI at Work Skills & Reskilling

AI is built to sound right, which is not the same as being right

Dan Klein, a Berkeley professor who builds AI systems, explains that today's AI is trained to sound convincing rather than to be correct. That is why it can hand you a confident answer that is quietly wrong, and why smooth writing is no longer proof that the facts under it hold. The skill that keeps its value is not producing the words, which the AI now does well, but being the person who checks whether they are true before they go out.

For youSet one rule this week, that nothing the AI writes for you leaves your hands until you or a second tool has checked the specific facts, names, and numbers in it.

Source: The AI Report ↗

AI at Work Skills & Reskilling

You can hand off the writing and still sound like yourself

Writing a post or a work update used to mean staring at a blank box until the words came. Now you can point an AI at a few things you have already written, answer a handful of its questions in your own rough words, and have it learn how you actually sound, then draft in that voice for you to trim. The writing gets handed off, but the voice and the choice of what to say stay yours.

For youGive Manus (manus.im, free tier) three things you have written and one messy paragraph of your real opinion, then judge its draft on a single question, does this sound like me, and keep editing until it does.

Source: The Rundown AI ↗

Skills & Reskilling AI Governance & Regulation

A wide open field is forming around making AI understandable

Right now even the people who build AI often cannot say why it gave a particular answer. A small but growing group is working to change that, building ways to read an AI's inner workings like a mechanic reads an engine, so it can be checked, corrected, and trusted. One founder in the field reckons only a few hundred people work on it full time, which for anyone choosing what to study or build is a rare open door.

For youIf you are early in your career or picking a thesis topic, spend an hour on Goodfire's public research pages (goodfire.ai/research) and notice how young the questions still are, because a field this thin on people is where a newcomer can matter fast.

Source: The Neuron ↗

The Human Signal.