Three futures for work, and the path that holds in all of them

Nobody can tell you what work looks like in 2030. I have spent 30 years in technology and nine working with AI, and I can't. Neither can the economists, the consultancies or the AI companies, and their forecasts disagree wildly.

An MIT economist expects AI to add about 1% to the US economy over ten years. One AI company's most extreme scenario has the economy a third larger by 2030, with knowledge workers paid less or out of work. Everything else sits somewhere in between.

The evidence so far is calmer than the headlines. Yale and Brookings looked across the whole US job market and found little sign of change since ChatGPT launched. Stanford looked closer and found that workers aged 22 to 25 in the most exposed jobs saw a 13% relative fall in employment, while experienced workers in the same jobs held steady.

Calm on the surface, movement underneath.

So I stopped asking which future will arrive. The better question is what plan holds up in all of them.

Three futures

In September 2026, economists at Anthropic modelled the US economy in 2030 under three scenarios. It is one model from an AI company, so I read it alongside the sceptics. But it turns a wide range of opinion into three pictures you can plan against.

Modest. AI feels like the arrival of the internet. The economy is 1.6% larger than it would be without AI, close to the cautious MIT view. Pay for knowledge work changes little, and unemployment stays within its normal range.

Substantial. AI can do half of all knowledge work, though most of it is still done without AI. The economy is 8.3% larger. Pay for desk-based knowledge work is flat, while pay for hands-on work such as trades and care goes up. When the same researchers surveyed 10,980 Americans, the typical answer pointed here.

Extreme. AI does nearly all knowledge work on its own. The economy is 32.4% larger, knowledge workers' pay falls by more than 10%, and unemployment rises beyond typical recession levels. Some of the economists who reviewed the model read this one as a thought experiment.

The numbers matter less than the shape. In every future the economy grows, and in the faster ones the pressure lands on desk-based knowledge work. What changes between them is how fast it arrives, and how long you have to adjust.

That is also how your role changes. In the modest future a development manager keeps the title and gets new tools. In the substantial one the title shifts, perhaps to something like an agent orchestrator: fewer people, more agents, and a job that is mostly deciding what gets built and checking it. In the extreme one the new roles exist, but there are fewer seats, and they go to people who already made the shift.

What holds in all three

The research disagrees on speed and size. It agrees on more than you would expect.

• Tasks shift before jobs disappear. A job is a bundle of tasks. AI leaves some alone, helps with some, takes over others and creates new ones. The useful question is which of your tasks are moving.
• Hands-on and care work holds up. The pressure is on desk-based knowledge work, and it shows up first in entry-level roles. Experience is holding its value, for now.
• Owners gain more than earners. Workers take about 60% of income today. The model has that share at 59.4%, 56.1% and 45.2% across the three futures. Even the cautious MIT economist expects the gap between capital and labour to widen.

So the plan starts in the same place whichever future arrives: move your week toward judgement. That is easy to say. The harder part is how.

The path toward judgement

If you are mid-career, you already have judgement. That is not the problem. The problem is that your week is still full of production, because that is what the job description and the timesheet ask for. Your judgement is invisible, sometimes even to you. And your income is tied to the production, so handing it to AI feels like handing over your pay.

1. Sort

   Know which of your tasks are production (output AI can now produce), assisted (you do it, faster with AI) and judgement (it needs you, your relationships or your call). I've written up a way to sort your own week.

2. Hand over and check

   Let AI draft. Your job becomes knowing what is wrong with the draft. I learned the hard way that neither people nor AI are good at judging the quality or security of what they built themselves, so the checking is real work.

3. Name your judgement

   Work out what people rely on you for, and say it plainly enough that someone could buy it.

4. Own the outcome

   Judgement gets paid when your name is on it and you carry the consequences. This is where the role changes, which is why the role change comes last, not first.

5. Price the judgement

   A record of sound calls in one field is rare, and rare things set their own price. Tie your income to outcomes and accountability, not hours.

Step 5 is where supply and demand come in. "Sound judgement" in general is not scarce, because nobody can check it. What is scarce is specific: the person who has taken this kind of company through this kind of problem. The narrower and more visible your record, the more bargaining power it gives you.

When hours stop describing the work

Picture a fractional CFO, sold to a growing business as half a day a week. On the invoice, that is the job. In practice, the business leans on them for the calls nobody else can make: whether the cash will last, which risk to take, what to tell the board. They carry that accountability all week, and they spend unbilled time on it because they care how it turns out.

The most valuable thing they do in a month might be one conversation that changes a decision. It rests on 25 years of experience, and no timesheet shows it. As AI takes on more of the production work around them, the hours shrink further while the accountability stays.

That is steps 4 and 5 in practice. The way through is to describe the role by what they are accountable for, agree in writing what they are not accountable for, and price the role, not the days.

Common questions

Steady judgement

You can't pick the future. You can pick where your week goes, what you are known for and how you are paid for it. Those hold up in all three.