Strategy is still a choice about where to compete and how to win. What AI has changed is the second half. For twenty years, a large share of mid-sized companies won because their product was better and hard to copy. Closing that gap used to take a competitor two or three years and a team of forty engineers. Today a rival with a clear specification and a handful of people working with agents can rebuild most of the visible product in a quarter. A strategy whose answer to “how do we win” is “a better product” now has a shelf life, and it is shorter than most strategic plans.

Value moves to the constraint that remains

The most useful frame I have found for this is in Reshuffle, Sangeet Paul Choudary’s book on how AI reorganises the knowledge economy. His argument is that organisations are built around a few constraints: expertise is scarce, someone has to carry the risk, and people with different interests have to be coordinated. When a technology removes one constraint, value does not disappear. It migrates to the constraints that are left.

AI removes a good part of the first one. Writing code, drafting a contract, building a financial model or producing a market analysis is no longer scarce. I covered what this means for software in build or buy software when code costs almost nothing. The same economics apply to competitors: the features you spent five years building are now a specification someone else can write after a few calls with your customers.

What does not move with it is everything that is not made of knowledge. Physical assets. People and the trust placed in them. The ability to make a whole system work together. That is where strategy has to look now.

AI strategy is not strategy

Most of what is published under “AI strategy” is an adoption plan: which use cases to pilot, which platform to buy, how to train staff. It is useful work, and it is not strategy. Every competitor has access to the same models at the same price.

Choudary illustrates the trap with the shipping container. Ports that used the container to load faster gained a little. The ports and companies that won were the ones that redesigned the system around it: a single contract across sea, rail and road, predictable transit times, and manufacturing that could run with less inventory because delivery became reliable. Singapore did not win on crane speed; it won on being the most reliable place to coordinate trade. Using AI to do the same tasks faster is the faster crane.

The test I use is simple. If the strategy document would read the same with a competitor’s logo on the cover, it is an operating plan. Strategy is the part that only works for you, because it rests on something you have that others cannot get quickly.

Physical assets: what cannot be downloaded

Not every company is a technology company, and the ones that are not are often better placed than they think. A network of well-located stores, a warehouse next to the motorway, a plant with environmental permits, a hospital licence, a fleet, guaranteed access to a raw material or to energy: none of this can be generated by an agent, and much of it takes years and regulatory approval to replicate.

As the digital layer of every sector becomes cheap to copy, scarce physical assets gain relative value. The logistics company whose differentiation was its routing software loses ground; the one that owns the docks and the slots keeps it. The retailer whose edge was its e-commerce site faces a hundred equivalent sites; the one that controls the best locations and the supplier relationships does not. AI itself runs on atoms: data centres, power and grid connections are among the scarcest assets of this decade.

The strategic question is which physical assets the business controls, how long it would take a competitor to obtain them, and whether the company is investing to secure more or treating them as a cost to minimise.

People: judgment, relationships and someone who answers

When expertise stops being scarce, what becomes scarce is the person who decides well in situations nobody has seen before, and the person the customer trusts. Those are different things, and both matter.

The first is judgment. Models are good at problems that resemble their training; they are weak on genuinely new ones. A company whose people can frame and solve problems that do not yet have a playbook adapts faster than one whose people execute procedures, because the procedures are exactly what AI absorbs.

The second is the relationship. A sales force that knows the buyers, an account manager the customer calls before making a decision, a partner who recommends you: none of this is produced by a model. When building costs almost nothing, the bottleneck becomes reaching the customer and being trusted with the decision. Companies that treated sales and customer service as a cost to cut will find they cut the asset that matters.

The third is responsibility, Choudary’s second constraint. Customers pay a premium to hand a risk to someone else: a certified process, a signature on a regulatory filing, a contractual commitment on accuracy or availability. An agent can produce the output; it cannot carry the liability. Businesses whose value sits in answering for results, in payroll, compliance, clinical processes or financial reporting, are more defensible than their software suggests.

All three depend on attracting and keeping a small number of people. Talent density, culture and the reasons a good person stays become strategic variables, not HR topics.

The system: who coordinates

The third constraint, coordination, is where Choudary sees the largest shift. AI lowers the cost of aligning people, information and resources across organisations. The winner is the company that uses that to become the place where the system is coordinated: the one that sees demand, supply and execution together and makes them work as one.

That position rests on two things. The first is data produced by operating: not data you bought, which a competitor can buy too, but the record of what happened when your business took part in thousands of real decisions. At Nextail, the hard part to replicate was never the interface; it was years of sales history across hundreds of stores and what it taught us about how retailers really decide. The second is learning speed: how quickly a signal from the market becomes a change in the product, in prices or in how the company sells. It is an organisational property, not a technical one, and two companies with the same tools can differ several times over in cycle time.

What stops being an advantage

A long feature list stops defending anyone, and so does a roadmap whose main argument is that it is twelve months ahead. The number of integrations a product supports used to be a barrier; agents now write one in an afternoon. A large team doing knowledge work that AI can absorb stops being a capability and becomes a cost line. First-mover advantage built only on product lasts until the second mover reads your public documentation.

None of this means product quality stops mattering. A poor product still loses. A good product is now the entry ticket, not the reason customers stay.

What changes in how a company plans

The test for any plan becomes: if a well-funded competitor rebuilt our product in ninety days, why would customers stay? If the answer is the product, the plan assumes a durability that no longer exists. If the answer is in the physical assets, the people or the coordination of the system, those are the assets to measure, protect and invest in.

The budget follows. Less goes into widening the feature set; more goes into securing scarce physical assets, into the commercial relationship, into keeping the people who judge and answer for results, and into the data and processes that let the company coordinate its system better than anyone else. The review cadence has to tighten too: a strategy reviewed annually against competitors who can change their offer every quarter is reviewed too late.

The question to ask

Take your current strategy and remove every sentence that depends on knowledge being hard to produce. What is left is your real strategy. If very little is left, that is the most important finding of the year.