Forecasting AI from a 20-minute test drive

You know the routine. The dealer hands you the keys and you get 20 minutes. A loop round the industrial estate, a roundabout, one straight bit of dual carriageway, a speed bump. Then you sit back down in the showroom and try to decide whether you could live with that car for the next 3 years.

Ask someone who has only done that loop in an electric car how far one goes on a charge, and you'll usually get a worry.

Ask someone who has driven one every day for a year, through a winter, and you'll get a number. Usually a bigger one.

Same car. Same battery. Different miles.

Research published this week found the same pattern in how people forecast what AI will do to their own jobs. It also says something awkward about where most senior professionals are forming their view.

Let's dive in.

On 16 October, one of the large AI labs published a survey comparing its own staff with the general public. The public side was fielded by Morning Consult in August: 10,980 US adults, weighted to be nationally representative. The staff side was an internal poll of 1,334 people.

Both groups were asked how much of their current work AI will be able to do by 2030.

US workers said 28%, on average. The lab's staff said 75%.

At the far end the gap gets wider. More than half of the lab's staff expect AI to be able to do at least 80% of their current work by 2030. Among US workers, it's 5%. About 1 in 20.

(AI lab research note, "The AI expectations gap", 16 October 2026; US survey fielded by Morning Consult, August 2026)

You'd expect 2 groups that far apart to disagree about how AI affects jobs. They mostly agree.

In both groups, the more of their own work people expect AI to take, the higher the unemployment they forecast. Every extra 10 percentage points of work affected went with 0.4 points more forecast unemployment. US workers who expect AI to do 80% or more of their job forecast 12.2% unemployment by 2030. Respondents were told the current rate is about 4.2%. The researchers note 12.2% would sit above the peak of the financial crisis.

Same reasoning. Different inputs.

The researchers trace the gap to one thing: how much each group has actually used the tools. 59% of the lab's staff say AI gets a task done in a quarter of the time or less. Among US adults who use AI at work, it's 26%.

A companion report on the lab's own workforce shows what that daily use looks like once it reaches hiring. The share of junior technical hires fell from 27% to 9% since early 2025. And when hiring managers scored technical candidates, comments on autonomy and judgement grew fastest, while domain expertise took a smaller share of the assessment.

One manager put it plainly: "I expect models to continue to be jagged, so for my team I want them to be fungible across domains."

(AI lab research note, "How employment is changing", 16 October 2026)

Put simply, the people with years of daily use expect far more change. And they're already hiring for it.

Here's where I think a lot of Directors misread their own position.

Most senior people I speak to have a view on AI and their role. It's usually calm. Some of the work will change. The judgement-heavy parts won't. There's time.

That view might be right. Now ask where it came from.

For most corporate leaders, direct experience of AI looks something like this: a licensed chat tool. A pilot in a neighbouring function. A demo at the offsite. A summary of a meeting you were in anyway. Someone more junior showing you something on their laptop.

That's the 20 minutes round the industrial estate. Enough to form an opinion. Nowhere near enough to know how far the thing goes.

The survey suggests your reasoning is probably sound. If you believed AI could do 75% of your work, you'd reach the same conclusions as the people who do believe it. The weak point is the input. You're forecasting from the test drive.

The second risk is time.

The same survey asked how long someone displaced by AI would take to find work in a new occupation. The median US adult said 7 to 12 months. The lab's own staff said 12. For comparison, the survey put normal times at about 3 months to find a new job.

So the group with years of daily use expects more change and a slower way back.

If they're even half right, the gap between needing a second option and having one runs to several quarters. That's the time it takes to build one deliberately. It's also the time you don't get once the decision has been made for you...

Back to the electric car.

You don't get an honest sense of range in the showroom. That comes from living with the thing. On the motorway, in January, with the heater on, on a morning when you're already late.

Your own exposure works the same way. You can't read it off a survey, this one included. You find out by putting a capable tool against the hardest real parts of your week and watching what happens.

That's uncomfortable for a senior person to do. It's much better done now, privately and on your own terms, than discovered later in someone else's restructure deck.

So take the long drive. This week:

1) Pick something you're proud of. A piece of work from the last 6 months that took real judgement. A forecast, a board paper, a commercial proposal, a reorganisation plan. Use a version that's safe to share, and stay within your employer's rules on data.

2) Give it 2 hours. Ask a capable AI tool you're allowed to use to produce a first version of the whole thing, starting from what you started with. Then mark it up. Where it was good. Where it was wrong. And where only you would have known it was wrong.

3) Re-forecast, then show someone. Write 2 numbers: the share of your week you thought AI could do before the drive, and the share you'd put on it now. Then take the marked-up version to one person outside your organisation whose judgement you trust, and ask which part of your work they'd still pay for.

That third step is the one that turns a private shock into an option. The parts only you caught are your judgement. The person outside the building tells you whether it travels.

None of this means leaving. It means forecasting from real miles while you're still employed and the timing is yours.

This is the work Zelova is built for. The guide looks at your actual role, helps you see which parts are exposed and which parts of your judgement other organisations would pay for, and turns that into a plan with more than one outcome in it. Built while you're employed, so the choice of terms stays with you.

It's a career strategy service, still being built. If any of this feels close to home, the waitlist is open at zelova.ai.

If you had to put a number on it today, how much of your current week could AI do by 2030? And how much of that number comes from actually using it?

Reply and tell me. I'd like to hear your version.

David