“Adopt AI” is a goal from 1899
It is 1899. A factory owner buys an electric motor. Until now, a single steam engine has driven a long shaft beneath the ceiling, with leather belts running from it to every machine. The owner puts the electric motor exactly where the steam engine stood and connects it to the same shaft. The next day, the workers return to the same factory floor and do the same work in the same way. Only the power source has changed.
In autumn 2026, a leadership team sits down to set goals for 2027. The slide says “AI first”. Someone suggests introducing an AI assistant in customer support. Someone else wants to measure how many people use AI tools. Everyone nods. I have seen the same item in the goals of almost every company whose plans I have looked at this year. Not once did it say who would be doing what differently by June.
In this article, I’ll show why a goal like this ends the day IT hands out AI assistant licences, and how to write it so that by June, someone is doing different work from today.
Factories paid for electricity for forty years before productivity improved
The first public power stations started supplying electricity in New York and London in 1881. Eighteen years later, in 1899, electric motors powered less than 5% of machines in American factories. It was not until the early 1920s that they powered more than half of all machines, and only then did manufacturing productivity rise.
Economist Paul David described this in his 1990 paper The Dynamo and the Computer, which asked why companies were not seeing the benefits of the computers they were investing in. His argument applies equally to AI.
Yet the electric motor worked from day one. For forty years, factories simply used it as a better steam engine: one large motor, one shaft, belts running to every machine, and a layout dictated by how far the belts could reach. Productivity improved only when each machine got its own small motor. The overhead shaft disappeared, and machines could be rearranged to follow the sequence of production steps.
David calls this fixation on the existing setup (functional fixity). The factory owner changed the power source but kept the factory floor arranged around the shaft. They were paying for electricity while producing goods in the same way as they had with steam.
An “adopt AI” goal is complete the day IT hands out licences
Goals such as “we need to be AI first”, “we need to be AI ready” or “we need to drive AI adoption” are measured by the number of licences for AI assistants such as ChatGPT or Copilot, or the share of people who opened one of these tools during the month. A goal like that is complete the day IT hands out the licences.
The 2025 MIT NANDA study The GenAI Divide drew on interviews with 52 companies, a survey of 153 managers, and a review of more than 300 publicly documented corporate AI implementations. It found that 95% of companies were not yet seeing any measurable impact on their financial results from their investment in generative AI, while just 5% of pilots were delivering millions of dollars in value. The authors identify the main barrier: the tools neither learn from how the company operates nor fit into people’s day-to-day work.
Those 95% of companies ended up like the factory with an electric motor connected to the old shaft. In practice, a salesperson gets a tool that summarises every call, but still spends the evening filling in the same fields in the sales system by hand, because those are the only fields used to build the report for management. A meeting is automatically transcribed, but the assistant still writes up the minutes in the old template afterwards, because no one has changed it and management still wants the minutes in that format.
Factories had forty years to redesign their operations. You have a few.
The factory owner of 1899 could afford to buy a motor and leave the factory redesign to their son. A company setting goals for 2027 has one year to make the equivalent change, because those goals will be reviewed again in December 2027. So the goal needs to spell out how work will change from the moment it is written.
A well-written goal answers four questions:
- which tasks will no longer be part of the role, and when;
- which tasks people will keep and what new responsibilities they will take on;
- who is responsible for AI’s output and how they will check it;
- what people need to learn, and by when, to handle their remaining and new responsibilities.
Rewritten this way, a goal to “introduce AI into customer support” looks different. From June, AI handles routine customer queries first, and people deal only with the cases it passes on to them. Two members of the support team move into onboarding new customers. By the end of February, they learn the product in depth from their onboarding colleagues. From March, they run introductory training sessions on their own.
The head of support is responsible for the accuracy of AI’s answers and checks a sample every week. Every part of the goal has an owner and a deadline. The head of support and the two team members learning the product all start in January. In December, the goal is assessed against what people are actually doing.
Open your list of goals for 2027
Take your list of goals for next year. For every goal that mentions AI or automation, ask who will be doing something differently by June, and what that change will be. If the answer is “they will use a tool”, the goal describes no change in the work itself. Rewrite it before you approve it. Otherwise, by December 2027, you will have paid for a year of licences while people are still doing the same work as today.
How many of your goals for next year would pass that test? And who in your company can say today which tasks will disappear once AI is introduced, and which will remain with people?
