Quiet demotion

A senior engineer at a small practice has written every structural inspection report it has issued for twelve years. This winter the firm starts a trial - an AI tool turns her site notes and photos into a first draft, and she checks it. The directors say it will free her for higher-value work. She is still waiting to find out what that is.

The inspection report is where her professional judgement shows: which crack is movement and which is shrinkage, and what to tell a client who really wants good news. Checking a draft still takes judgement, but nobody sees it. The report used to show her thinking. Now it shows the tool's, with her corrections, and her signature.

On Odd Lots Francis Fukuyama describes thymos, the desire to have your dignity recognised, and argues that money alone cannot satisfy it. Pay matters partly because it tells you that somebody considers your work worth paying for. Take away the part of the job that earns someone recognition, and keeping their salary may not be enough.

The result is quiet demotion: AI lowers a person's standing while their job title is unchanged.

Why some tasks matter more than others

Fukuyama explains why losing recognition at work hurts. But which parts of a job give someone that recognition?

Game design offers one way to think about this. Keen gamers put considerable effort into a game without being paid. Paula Gomes uses self-determination theory to explain why: good games give people a sense of competence ("I am getting better"), autonomy ("I choose how") and relatedness ("others see me").

For Fukuyama, dignity comes from others recognising that your work has value, and pay is the signal. Standing is the same recognition inside the firm: who asks your opinion, whose sign-off counts. It comes from two of the above needs - competence says "I am good at this", relatedness says others know it. Autonomy is how a rollout protects or damages both.

How a rollout drains standing

Start with competence. AI removes the work that develops skill. In an Anthropic trial of 52 developers learning a new library, the group with AI help scored 50% on the comprehension quiz against 67% for the group without.Judy Hanwen Shen and Alex Tamkin, How AI assistance impacts the formation of coding skills, arXiv, 2026. Matt Pocock puts it bluntly: AI has eaten the routine work that used to train juniors. His fix: "Hire juniors. Give them work that matters. Watch them fail. Pick them up." A junior who only checks AI output never builds the judgement to check it properly. That weakens the firm's training pipeline and the junior's standing. The consequences emerge years after the time was saved.

Autonomy goes next, usually by mandate. Meta tracked the AI token use of 85,000 staff on an internal leaderboard, and similar boards ran at Microsoft and Salesforce, where the top users produced throwaway work. An expected prize for using a tool is the textbook way to kill interest in it. Deci, Koestner and Ryan's meta-analysis of 128 experiments found that paying people to do a task made them less likely to do it for fun afterwards.Edward L. Deci, Richard Koestner and Richard M. Ryan, A meta-analytic review of experiments examining the effects of extrinsic rewards on intrinsic motivation, Psychological Bulletin, 1999. Positive feedback made it more likely.

Relatedness goes quietly. On another Odd Lots episode, social media consultant Rachel Karten describes bosses who run their staff's work past Copilot and trust the model over the humans who did it. One team now pastes its own ideas into a mock Copilot screen so the boss will take them seriously. In Hollywood, people hide their AI use. Hayes Davenport, who wrote for Family Guy, says using AI on a production draws public pushback, so people keep it to themselves. Recognition used to follow visible effort, and cheap output has already weakened that link.

Headcount is not standing

The obvious objection: companies that use AI heavily are hiring more people, not fewer. Ramp found the most aggressive adopters among 22,000 US firms grew their white-collar workforce 10.2% in two years.

But headcount is not a measure of standing. A working paper by Bharat Chandar and Bouke Klein Teeselink, using job records from 41 countries, finds that adopting firms hire more seniors while the junior share shrinks.Bharat Chandar and Bouke Klein Teeselink, How Does AI Change Labor Demand? Evidence from 41 Countries, SSRN working paper, 2026. Legal roles show the largest fall, 4.9 points. The people stay, but the shape of their roles change.

Standing can also move. In many firms the graduate who can make the tools work is suddenly visible to the directors. The two effects differ - the senior engineer faces quiet demotion, while the graduate who only checks AI output may never gain the skills required to be promoted at all.

Start with one inspection report

Fukuyama borrows a warning from Tocqueville: revolt follows reform, because "expectations were rising faster than the regime's ability to accommodate these new expectations." A company-wide AI launch does exactly that. It promises everyone higher-value work and frightens them about what they will lose, before anyone has seen the tool do anything. It makes sense to narrow the scope, ship one change people can see, then widen it.

How you make that change matters as much as what it is. Gagné and Deci describe three things that help people take on a change as their own: a real reason for it, an admission that the task may be dull, and choice over control.Marylène Gagné and Edward L. Deci, Self-determination theory and work motivation, Journal of Organizational Behavior, 2005. When managers were trained to work this way, their staff came to trust the organisation more.

Put together, that gives you a test. Pick one process and ask the person who does it which steps tell them they matter. Compare their answer with yours. For the engineer, it will probably be the judgement on each crack and the conversation with the client. Leave those with her. Hand the AI the rest, but make sure she can check the output.

None of this tells you whether people will accept the change. Some take a smaller job willingly. Philip Su, a former distinguished engineer at Facebook and OpenAI, asked to move down from IC9 to IC7. Past a certain level the job had become steering teams, sitting in meetings and handling unhappy colleagues, and he did not want it. He called it "realism, not impostor syndrome". He saw that the job had changed shape, but quiet demotion often goes unseen.

Before your AI rollout, add the missing column. For every job the pilot touches, note four things now and check them again in three months: who gets asked for an opinion, whose sign-off counts, who faces the client, and whose name is on the work.