Companies pushed staff to use AI. It backfired — and workers are pushing back

Information Technology 4 min 75 sources

The corporate order to 'use more AI' is misfiring inside the firms that gave it, software layoffs are feeding a political backlash, and the chipmakers underneath the boom keep breaking records.

Key takeaways

Companies that ordered staff to use AI more are finding the push backfired; in software, where AI now writes much of the code, layoffs and deskilling fears are feeding a political backlash — including a poll where 69% backed making AI firms share half their stock with the public.
The chipmakers beneath the AI boom keep breaking records: TSMC's June sales jumped about 68% and it is building new plants to keep up.
A Brazilian court forced Microsoft to restore a gamer's suspended library — a small but early ruling that a purchased digital collection is the buyer's to keep.

The productivity push meets a wall

Over the past year, a lot of companies gave their staff one instruction: use AI more. This week the Financial Times reported what many workers already felt — the push has largely backfired. [32] Told to lean on tools they didn’t pick, to hit targets set by people who don’t do the work, employees are learning to satisfy the mandate rather than gain from it.

The sharpest picture is in software engineering, one of the best-paid jobs in the US and now one of the most unsettled. Since ChatGPT arrived in 2022, more than 600,000 US tech workers have lost their jobs, by the count of the tracker Layoffs.fyi. [9] Unemployment among computer-science graduates hit 7% in 2024, up from 6.1% the year before, and nearly one in five was underemployed, according to the New York Federal Reserve. [9] US tech job postings fell 36% between 2020 and 2025. [9]

The tools are doing more of the actual coding. Google has said about 75% of its code is now written by AI. [9] So the human job is shifting — from writing code to checking code the machine wrote. One engineer who spoke to the Guardian, and asked not to be named, described his boss warning him to use AI more while he quietly refused where he could, to keep his skills, as he put it, sharp. [9]

That friction is now spilling into politics. A June survey of 1,690 Americans by the research firm Verasight found 69% support forcing AI companies to hand 50% of their stock to a public fund that would share the profits with everyone. [4] Senator Bernie Sanders has introduced a bill to build exactly that — a $7 trillion “AI Sovereign Wealth Fund” giving the public a 50% stake in the largest US AI firms. [4] The anger has a clear source: layoffs are climbing even as company profits rise. [4]

The hardware underneath is having a record year

While the AI rollout frustrates the people using it, the companies making the chips beneath it are booming. TSMC — Taiwan Semiconductor Manufacturing Company, which builds the chips for Nvidia, Apple, and most of the AI industry — said its June sales rose about 68% from a year earlier. [21] Revenue for the first half of 2026 reached roughly $75 billion, up 36%, which the company put down to demand for AI chips. [21][24] Its full second-quarter results land later this week. [24]

To keep up, TSMC is adding two “advanced packaging” plants in Chiayi, in southern Taiwan, a government minister said. [18] Advanced packaging is the step where separate finished chips are stitched into one module — the assembly stage that has become the real bottleneck for AI hardware, harder to scale than the chips themselves.

Apple is leaning harder on its own silicon too, and how it got there is a reminder that failed projects aren’t always wasted. Apple’s cancelled self-driving car, code-named Project Titan, never shipped. But the effort produced the Neural Engine — the part of Apple’s chips that runs AI on the device itself instead of in the cloud. [20] It first appeared in the 2017 iPhone and now anchors the company’s plans: according to Bloomberg’s Mark Gurman, Apple is rushing its next chip, the M7, out for early 2027, with a top-end version built to power a new Apple server holding up to 1.5 terabytes of memory. [20][16]

A small court win for people who thought they owned their games

A quieter story from Brazil is worth the ending. A gamer who goes by Ordo_Liberal on Reddit had their Xbox account permanently suspended by Microsoft after what the company called “unauthorized access” — losing an entire library of bought games with it. [35] Rather than start over, they sued. A Brazilian court ordered Microsoft to restore the account within 15 days or pay a fine of about $30 a day, plus roughly $400 in damages. [35] The sums are tiny; the principle is not. As games and films move to “buy” buttons that are really rentals a company can switch off, this is one of the first courts to rule that a purchased library is the buyer’s to keep. Brazil’s strong consumer-protection law made the case free to bring. [35]

And one more sign of where the work is heading: India’s Tata Consultancy Services, the country’s largest IT firm, said it plans to hire up to 8,900 “AI deployment engineers” and to buy AI startups. [22] The job title is the tell. The growth isn’t in writing software from scratch — it’s in wiring AI tools into other companies’ systems and keeping them running.

02 · Lesson · why it matters

The skill you'll need to check the machine is built by the work it's taking

Everyone agrees the job is becoming to supervise what AI produces — but the ability to supervise it well is grown by doing the work yourself, which is exactly the part being handed over.

Two instructions that don’t fit together

Read the week’s tech news closely and you find companies telling their staff two things at once.

The first is an order: use AI more. Hit the tool harder, ship faster, and expect a warning if you don’t.

The second is a reassurance, offered to the same anxious workers: don’t worry about the machine writing the code — your value now is in judging what it produces. Writing is being automated; reviewing is the future.

Both statements come from the same mouths. And they point in opposite directions.

Where judgment actually comes from

To see why, ask a plain question: how does anyone get good at spotting a wrong answer?

Not by reading about mistakes. By making them. You write the clumsy version, it breaks in a way you didn’t expect, and you spend an afternoon finding out why. The next time you meet that shape of problem, something in you flinches before you can even explain it. Expertise is mostly a private library of errors you have personally made and paid for.

The reviewer who catches the subtle bug in someone else’s code catches it because they have written that same bug themselves, at two in the morning, and lost a night to it. Take away the years of writing, and the library stops filling. You are left holding the job title “reviewer” without the thing that ever made a reviewer worth having.

The tool removes the training ground, not just the task

This is the quiet trap inside the productivity push.

The AI does the first draft. That is the visible gift — the hours saved, the output shipped, the number that climbs on the dashboard. What doesn’t show is that the first draft was where the skill was being built. Automate the doing, and you slowly starve the judging.

And the judging is the one job everyone agrees will matter more, not less. So the harder and more successfully you lean on the tool, the less able you become to do the single thing the tool cannot do for itself: know when it is wrong. The order to use it and the reassurance about your future both assume a competence the order is busy wearing away.

This is not only about code

It travels far past software.

The radiologist reading scans the AI has already flagged still needs the eye that says “that flag is wrong” — an eye trained by years of reading scans with no flag at all. The pilot watching the autopilot needs the instinct to take the controls in the one minute it fails, an instinct built on thousands of ordinary minutes of flying by hand. The junior lawyer’s dull document review was never only dull work; it was how the senior partner’s judgment got made.

Wherever a machine takes over the entry-level reps, it quietly saws off the lower rungs of the ladder that produced the experts standing at the top. The underemployed computer-science graduate isn’t only short a paycheck. They are short the plain, unglamorous years of work that would have turned them into the person who can check the machine — the person their whole field is about to need.

You are somewhere on this ladder too. Whatever task you have lately started letting a tool take the first pass on, you are also deciding what you will slowly lose the feel for.

Why no one is exactly the villain

It would be easier if someone were doing this on purpose. No one is.

The manager ordering more AI has an honest reason. This quarter’s output and this year’s costs are both real, both measurable, both what the company rewards. Skill decay is a cost that lands three or four years out, on someone whose judgment never formed — and it shows up on no balance sheet, in no bonus. The arrangement isn’t hiding the cost out of malice. It simply can’t see it, because the gain and the loss keep different clocks: one ticks by the quarter, the other by the decade.

That mismatch is also what the anger behind the wealth-fund proposals is reaching for. When the people collecting the gains and the people carrying the costs are not the same people, someone eventually goes looking for a lever.

What the machine can’t hand back

You cannot opt out of this. The tools are here, refusing them carries its own cost, and they are not going away.

But seeing the shape lets you hold the word productivity a little more loosely — because you now know the part of the ledger it doesn’t print. The machine can do the work. It cannot yet do the knowing-whether-the-work-is-right. And that knowing is the one thing you can only keep by doing some of the work the machine offered to take off your hands.

The engineer on the four-hour train, writing every line of his own game by hand to keep his skills sharp, isn’t being nostalgic. He has understood something his own boss hasn’t: the hands stay good only while they keep touching the work. That is not a small thing to be holding on to. It may be the whole thing.

03 · Lab · your turn

Ship Now or Stay Sharp

Rehearse the trade between leaning on AI for output now and keeping the skill to catch its mistakes later.

04 · Hope · carry this

Every generation has been handed a tool that did part of its work, and the people who kept their hands in the craft are the reason the tool was ever worth trusting. The engineer refusing to let the machine do all of it isn't falling behind — he's guarding the human judgment the whole system still quietly leans on.