Blog
People4 min read

What AI actually does to the people using it

Not what it does to headcount. What it does to the working day of someone who uses it, measured by an organisation with no product to sell.

A worker feeding sheets into a large industrial press in a factory.
PeoplePhotograph via Unsplash

Every conversation about AI at work runs immediately to job losses, and stays there. It is the most interesting question and the least answerable one, and while everybody argues about it a much more tractable question goes unasked: what happens to the working day of a person who actually uses this stuff?

That has been measured. The OECD surveyed 5,334 workers across 2,053 firms in manufacturing and finance in seven countries, and asked the people using AI what it had done to their performance, their enjoyment of the job, their health and how fairly they felt they were treated. It is the best dataset on this that exists, it was published by an organisation with nothing to sell, and almost nobody quotes it.

The headline, which is not the one you expect

4x
Across every indicator tested, AI users were more than four times as likely to say AI had improved their working conditions as to say it had worsened them.OECD survey of workers and employers · OECD, The impact of AI on the workplace

That is the finding. Not a small positive, not a mixed picture: a four to one ratio in favour, across job satisfaction, physical health, mental health and fairness in management, from the people doing the work.

The individual numbers are as strong. Eight in ten said AI had improved their own performance, in both sectors, against eight per cent who said it made it worse.

80%
of manufacturing AI users said AI had improved their own performance, and 79% in finance. In both sectors 8% said it had worsened it.OECD survey · OECD

The part that matters more than performance

Performance improving is unsurprising. What is genuinely surprising is what the same people said about how the work feels.

63%
of AI users in finance said AI had improved their enjoyment of work, and 54% said it had improved their mental health and wellbeing.OECD survey · OECD
65%
of AI users in manufacturing said AI had improved their physical health and safety.OECD survey · OECD

Read that second one twice. Roughly two thirds of manufacturing workers using AI said it had improved their physical health and safety. That is a different kind of measure from a productivity figure, and it is the one the OECD reports for manufacturing where it reports enjoyment and mental health for finance.

The people who use it are not frightened of it. They are frustrated that it has not gone further.

The counter-evidence, which belongs here too

A page that published only the good half would be doing what vendor research does. The same survey found real costs, and they cluster in one place: control.

31%
of manufacturing workers who are managed by AI said it had decreased their control over their own tasks, and 23% in finance.OECD survey · OECD

Managed by AI is the operative phrase. Using a tool improves the day; being directed by one takes something away. That is not a subtle distinction and it is the entire design question in any system we build: does the person operate it, or does it operate them?

On job security, the fear is real and smaller than the coverage implies.

19%
of finance workers were very or extremely worried about losing their job to AI in the next ten years, against 46% who were not worried at all.OECD survey · OECD

Roughly half of workers in both sectors are not worried at all. One in five is very worried. Both of those are true at once, and any account of this that reports only one of them is selling something.

The two things that changed the outcome

The most useful part of the survey is not the averages. It is what separated the workers who reported good outcomes from the ones who did not, because both of those things are decisions a business makes rather than facts it inherits.

  • Training. Workers who had received training reported materially better outcomes than those who had not, and were more likely to say AI had improved their working conditions. Employers overwhelmingly address skill gaps by retraining the people they already have rather than by hiring: 64% in finance and 71% in manufacturing.
  • Consultation. Just under half of employers who adopted AI consulted workers or their representatives beforehand. Those that did were more likely to report positive effects on both productivity and working conditions, and in manufacturing were less likely to report AI decreasing overall employment.

Neither of those is a technology decision. They are the two cheapest interventions available in any rollout and they are the two most reliably skipped, because they slow the launch down by a fortnight and nobody is measured on them.

What this changes about how we build

Three things, and they come straight out of the data above rather than out of a philosophy.

  • Put the machine on the part of the job that is unpleasant or unsafe, not on the part that is visible. The physical health finding is the strongest single result in the survey and it comes from exactly that choice.
  • Never let the system direct the person. A tool a person operates improves their day. A system that assigns and paces their work takes control away from a third of the people subject to it.
  • Budget for the training and the consultation before the build, not after it. They are what separates a good outcome from a bad one, and they are the first things cut when a timeline slips.

None of that is expensive. All of it is skippable, which is why it usually gets skipped.

Sources

02

New writing, when there is some

A short note when something is published: what it is about and a link. No more than a couple a month, and nothing else.

Start with
the diagnostic.

A scoping conversation costs nothing and ends with a straight answer about whether there is enough here to be worth doing. If there is not, we will tell you.

Book a scoping callinfo@skaiyon.com

Questions. Asked before every engagement.

Your data is yours. You can export it at any time, and it is exported to you in a documented format before any engagement closes. The software itself is licensed: we build it around your business, host it, and run it, and you pay monthly for that. If you'd rather own it outright, that's possible. It's a different kind of engagement and it's priced accordingly. Licensing keeps maintenance our problem rather than yours, which is why it is the default.

Two weeks' notice, either side. Your data is yours. You can export it at any time, and it is exported to you in a documented format before any engagement closes. The system stops running. If you'd rather keep it running, the ownership option is available at that point as well. Ending the retainer doesn't force you to lose what was built.

Typically four to eight weeks from signing to a working system, depending on how many tools it connects to and what state the data is in. A scoping conversation and a written plan come first, so the timeline is agreed before anything is committed to.

Access to the tools and data the system will work with, one person who can make decisions, and a few hours in the first two weeks while we map how work actually moves through your business. After that, very little. The point of the engagement is that it runs without your attention.

Usually. Most business software exposes an interface we can build against, and where one doesn't there's normally a way around it. Which connections are viable is settled in the scoping conversation, so you find out before committing rather than after.

Hosting region for a system we build for you is decided at the start of the engagement, not inherited from a default. If data residency is a hard requirement, raise it in the first conversation and we will tell you plainly whether we can meet it. This website is separate and its processing is already fixed: enquiries, quote requests, bookings and chat are handled by Supabase in Tokyo, Resend in Tokyo, Cloudflare, cal.com and Moonshot AI, which the privacy policy names individually along with where each one processes. Nothing from a client system is ever sent to the chat assistant.

That's the normal starting point, and mapping them is part of the work. Automating a process nobody has examined just makes the confusion faster, so we don't start there. The first phase establishes how things actually happen, as opposed to how they're supposed to.