Thinking

Why long-term strategy fails AI-enabled companies

AI has made your company faster at everything except deciding what to do next

Something strange has happened inside the companies that have taken AI seriously. The work has sped up, often dramatically, so that a team which used to need a quarter to ship a pricing page or a first version of a feature can now do it in a week or two, and yet the company as a whole doesn't feel any faster, because the strategy is still set once a year, the board pack still arrives once a month and the big decisions still wait for the next planning cycle.

My view is simple, and it's the argument of our new white paper: AI has made your company faster at everything except deciding what to do next, which means the plan itself has become the bottleneck. This post is the short version; the full paper, with the research, a composite example built from the studies and the cadence we recommend, is free to download.

The plan was failing before AI arrived

It's tempting to blame AI for breaking strategic planning, but the research says the patient was already unwell. In a survey of 156 senior executives reported in Harvard Business Review in 2006, Michael Mankins and Richard Steele found that only 11% strongly believed strategic planning was worth the effort, and that every one of those executives said they made strategic decisions continuously, whatever the planning calendar said. Companies that stuck to the annual cycle made an average of 2.5 major strategic decisions a year, while companies that planned continuously, issue by issue, made 6.1.

McKinsey found the same thing in a survey of 796 executives, where only 23% said their major strategic decisions were actually made inside the formal planning process, and the same two authors had already shown that most strategies deliver just 63% of the value they promise. Donald Sull and his colleagues then looked at 124 organisations for MIT Sloan Management Review and found that only 28% of the managers responsible for executing a strategy could list three of its priorities.

So before anyone had used a large language model, the long-range plan was believed in by a small minority of the people who wrote it, bypassed when the real decisions were made and understood by fewer than a third of the people delivering it. Companies kept doing it because, when the work itself was slow, a slow plan didn't cost very much.

What AI changed

That slack has gone. METR, which evaluates frontier AI systems, has found that the length of task AI agents can complete on their own has doubled roughly every seven months for six years, and Stanford's AI Index reports that AI went from solving 4.4% of real software engineering problems in 2023 to 71.7% in 2024, while the cost of running a model at the level of the original ChatGPT fell more than 280-fold in two years. Inside companies, consultants using AI in a Harvard and BCG experiment completed 12.2% more tasks 25.1% faster, and customer support agents resolved 14% more issues an hour.

Put those numbers next to an annual plan and the problem is obvious. A twelve-month roadmap written in October describes a company with last autumn's tools, last autumn's costs and last autumn's idea of what's possible, so initiatives planned for the third quarter get done in the first, hires that made sense in the budget stop making sense by spring, and the thing that's really holding the company back usually isn't on the chart at all.

The bottleneck moved

If AI is this powerful, why are so many companies not seeing the results? MIT's Project NANDA reported in 2025 that around 95% of organisations were getting no measurable return from generative AI, and pointed to a learning gap and to tools that didn't fit how the company's work actually ran, rather than to the models themselves.

Eliyahu Goldratt would have recognised the pattern immediately. The Theory of Constraints says that every system is held back by one constraint at a time, that an hour lost at the bottleneck is an hour lost for the whole system, and that an hour saved anywhere else is a mirage. Most companies have spent two years saving hours everywhere except the bottleneck, making writing, coding and analysis faster while leaving the way they decide, align and learn exactly as it was. I made the same argument about Goldratt's four technology questions in AI is necessary but not sufficient: AI removes the limitation of slow execution, but most companies are still living by the rules that slow execution created, including the rule that strategy is something you set once a year and then deliver.

The fear of losing the Gantt chart

I understand why leaders hold on to the long-term plan, because a Gantt chart gives a leadership team a feeling of control and a story it can tell a board. But if strategies deliver 63% of their promised value and most decisions happen outside the plan, that control was mostly an illusion, and everything leaders actually need from the chart, which is direction, accountability, visibility and confidence, can be had without predicting the next eighteen months.

Direction comes from a clear long-term goal, accountability from every piece of work having one owner and a due date, visibility from a live picture that updates as the work moves, and confidence from seeing, every week, that the thing most in the way of the goal is being dealt with.

The destination is fixed; the route unfolds

The cadence we've built Clearahead around keeps one thing fixed and lets the rest move. The goal is the target, written in the team's own words with an owner and a date, and the route is found by asking one question again and again: what's holding us back from this goal right now? The answers are the constraints, put in order by the team, and under each one sit a few moves with an owner and a due date. When a constraint clears, the team records what cleared it, the next one moves up and the question gets asked again, so that the strategy becomes the sequence of constraints you clear, and you only find out what that sequence is by clearing them.

The rhythm runs at the speed of the work. People tell AJ, our AI, what's changed in a sentence as it happens; AJ writes a brief every Monday covering where the goal stands and which decisions need someone; the team checks the order of its constraints every month and asks whether the goal is still right every quarter. I set out the whole model in The operating model for the AI era and the case for it in A loop, not a plan.

Goals, constraints, moves and progress

That's what Clearahead is for, and it comes down to four parts. Goals sit at the top of every workspace with an owner, a due date and the thinking behind them, so the destination stays in view. Constraints are what's in the way right now, in the team's order, found by AJ asking what's holding you back until the answers get past the symptoms. Moves are the few actions that clear each constraint, planned weeks ahead rather than a year ahead, each with one owner. Progress comes from people updating AJ in a sentence, AJ keeping the current reality of everything up to date and spotting when a constraint looks cleared, and the owner making the call. You can see how it fits together on How it works.

Questions people ask

Why does long-term strategy fail AI-enabled companies?

Because AI has made execution fast while strategy is still set once a year. The plan’s assumptions about cost, time and what’s possible go out of date within months, and decisions queue up behind the planning calendar, so the plan itself becomes the constraint on the whole company.

What should replace the annual strategic plan?

Keep a long-term goal fixed as the target, then ask what is holding it back right now, put the constraints in order, clear the top one with a few owned moves, and repeat. The strategy becomes the sequence of constraints you clear, reviewed weekly, monthly and quarterly.

Do companies still need a long-term plan?

They still need a long-term goal and the thinking behind it. What loses its value is the detailed eighteen-month roadmap, because it predicts a route that AI-speed work will change long before the plan is finished.

Read the full paper

The white paper goes much further, with the full research, a composite company that shows how a well-built plan comes apart inside a quarter, the weekly, monthly and quarterly rhythm in detail and a plan for your first thirty days. It's free, with no form to fill in, and you can read about it here or download it below.

Download The plan is the bottleneck (PDF)

Alex Durston
Written by

Alex Durston

Founder, Clearahead

Alex built Clearahead to make it easier for teams to find what’s really stopping them from reaching their goals, and then clear it. If you want to talk any of this through, connect on LinkedIn or drop him a line. He reads everything.