No. 011Ways of Thinking

Amara's Law

We tend to overestimate the effect of a technology in the short run and underestimate it in the long run, so plan for both the hype and the structural change.

We expect too much from a technology in the first years and too little over the long run.

To an executive

New technology usually disappoints this year and transforms things over a decade. Avoid big bets on short-term promises, but start the structural changes the long run will need.

To an engineer

Treat near-term AI claims sceptically and prototype before committing. Meanwhile invest in foundations that pay off whatever the timing: clean data, clear interfaces, good evals.

The idea

"We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run." The saying is widely credited to Roy Amara, a futurist who was president of the Institute for the Future in California, and is known as Amara's Law.

Its exact origin is uncertain. Quote Investigator, which traced the family of similar sayings, found a 1983 news article paraphrasing Amara on investors overestimating robotics and biotechnology in the short run, and later printed versions in his name, but no single original publication. A related form, about overestimating what can be done in one year, appears anonymously even earlier, in J. C. R. Licklider's 1965 book Libraries of the Future. So treat it as an observation that Amara made famous, not a law he proved.

Why it works

Early expectations run ahead of the work. A new capability is visible long before the data, processes, skills and integrations it needs exist. Short-term results disappoint because the surrounding system has not changed yet.

Disappointment then hides the slow change. After the first wave fails to deliver, attention moves on, while costs fall, tools mature and practices spread. The long-run effect arrives through many small changes that nobody announces.

Long-run effects are structural. The lasting impact is rarely the original use case. It comes from reorganised work, new business models and changed expectations, which are hard to imagine in advance.

Where it applies

AI strategy and funding, to resist both the board's urgency and the sceptic's dismissal.
Vendor selection, where roadmaps promise next quarter what the market may deliver in years.
Enterprise architecture roadmaps, which must prepare data, interfaces and operating models for a change whose timing is unknown.
Stakeholder conversations, to reset expectations after a pilot disappoints without abandoning the direction.

How to apply

Separate the horizons. Write down what you expect this technology to change in one year and in ten, and fund them differently.
Make short-run bets small and testable. Prototype on real tasks, measure, and scale only what proves itself.
Invest in no-regret foundations. Clean shared data, machine-usable interfaces and evaluation practice pay off whenever the long run arrives.
Watch for structural signals. Track falling costs, competitors' operating changes and new skills, not only headline demos.
Revisit on a schedule. Review the long-run view yearly so that disappointment does not quietly end the preparation.

Example scenario

A logistics group approved a large programme to replace its customer-service centre with AI agents within a year. Six months in, the agents handled only simple enquiries, because order data was duplicated across five systems and none of them could be updated through an interface. Leadership began to talk about cancelling AI work altogether.

The enterprise architects reframed the plan with Amara's Law. They cut the short-term target to two well-measured use cases, and moved the rest of the budget to foundations: a single order record, APIs over the core systems and an evaluation pipeline. Two years later new models arrived that could use those interfaces, and the group expanded automation quickly, while competitors who had cancelled were starting from scratch.

Questions to ask

Which of this year's AI promises are we funding as if they were already proven?
What would our architecture look like if this technology became ordinary in ten years, and what are we doing now to prepare?
Which small, cheap bets keep us learning without committing us to the hype?

When not to use it

Mature, well-understood technologies, where short-term forecasts are usually reliable.
Specific delivery estimates, which need evidence from the work, not a general tendency.

How it misleads you

The law can excuse any failed prediction: hype that never pays off can always be called "long run". It does not say which technologies will matter, only that timing is commonly misjudged. Use it to balance short and long horizons, not to justify every bet or dismiss every doubt.