AI Product Strategy and What Actually Changes

A surprising amount of AI strategy is still a list of capabilities. We could add a copilot here, automate this workflow, introduce an agent there, generate this faster, personalise that. All useful possibilities. A list of things technology can now do is not yet a strategy.

The more useful question is what AI genuinely changes about product strategy, and what remains stubbornly true underneath it.

I think three things have changed enough to matter. The cost of exploring an idea has fallen. The ability to create is spreading beyond the traditional Product and Engineering system. Increasingly, the things we create can act rather than simply wait for a human to use them.

Those are significant shifts. They do not make customer attention more abundant. They do not make trust easier to earn. They do not remove the cost of owning something once it exists. They also do not remove the need for somebody to make a judgement and live with the consequences.

That distinction matters because it changes where product strategy needs to become more rigorous.

The cost of exploration has changed

Ideas are becoming much cheaper to make tangible. A prototype that might once have required weeks can sometimes exist in a day. Analysis, code, content, research synthesis and variations can all be produced faster. Not everywhere, not perfectly and certainly not at zero cost, but enough to change the economics of exploration.

That should be good news. It means more assumptions can be tested before they become expensive commitments. Teams can put something real in front of customers rather than debating an idea in the abstract. They can learn earlier, compare more options and discover that an idea is weak before a large amount of organisational energy has attached itself to it.

Cheaper exploration only creates an advantage if we remain willing to throw things away.

A prototype can have done its job perfectly without ever becoming a product. Something built to answer a question does not automatically deserve a roadmap, a launch plan and a permanent owner simply because it now exists.

This is where I think cheap creation can easily mislead us. The cost of producing an option is falling much faster than the cost of owning one. Once something becomes part of the product or the organisation, it starts asking for maintenance, support, security, documentation, monitoring, customer communication and future decisions. It may create dependencies that did not exist before. Someone has to remember how it works. Someone eventually has to decide when it should change or disappear.

Cheap to create is not the same as cheap to keep.

The opportunity AI creates is therefore not simply to build more. It is to make learning cheaper while keeping commitment deliberate.

Creation is moving beyond the roadmap

For a long time, Product and Engineering sat on a relatively scarce route to production. Ideas entered a system, were prioritised, designed, built and eventually released. That system was often slow and frustrating, yet the scarcity did some filtering for us. Not every idea could become real because there simply was not enough capacity.

That is changing.

Someone in Sales can build an agent for proposals. Operations can automate part of a customer workflow. A designer can make something functional rather than merely visual. Engineering can produce more with the same team. A founder can test a working proposition without first assembling a traditional product organisation.

I think much of this is positive. Useful creation should not belong exclusively to one function.

It does, however, change the leadership problem. Product cannot respond by becoming the approval queue for everything an organisation is now capable of creating. That would simply replace an old production bottleneck with a new governance bottleneck.

The more important question becomes what actually needs to be governed.

We need to be clearer about the outcome a capability exists to create, the boundaries within which it can operate, who owns the consequences once it exists, and what would cause us to change or stop it later.

The roadmap once performed part of this role indirectly because it allocated scarce production capacity. As creation becomes easier and more distributed, scarcity can no longer do as much of the governing for us.

Judgement has to become more explicit.

Autonomy raises the cost of fuzzy intent

There is another shift that matters even more than speed.

We are moving from software that helps people act towards systems that can increasingly act on their behalf. They can decide, communicate, modify information, trigger work, recommend actions and in some cases transact without asking a person at every step.

That changes the quality of the specification we give them.

“Improve engagement” was always an incomplete objective. So was “reduce handling time” or “increase conversion”. Human teams often compensated for that incompleteness. They carried context that was never written down. They noticed when pursuing a metric literally would create an absurd result. They asked questions, exercised judgement and interpreted what the organisation probably meant rather than only what it had formally said.

The more autonomous execution becomes, the less safe it is to rely on that invisible layer.

Incomplete intent becomes more expensive when the system can act on it.

Defining an outcome is no longer enough. We also need to define what must remain true while that outcome is pursued. If an assistant is reducing customer effort, perhaps trust must not fall. If an agent is resolving more service requests, perhaps certain decisions must still be reviewed by a person. If a system is improving conversion, perhaps it must not do so by making cancellation deliberately harder.

Those are not compliance checks to bolt on afterwards. They are part of defining what success actually means.

This is one of the places where I think Product has an increasingly important role. Product should not own every AI policy or approve every autonomous action. Its value comes from working at the intersection of customer value, business outcomes, behaviour and unintended consequences.

The technology is making that work more important.

What has not changed

AI changes the cost and distribution of execution. It does not change the basic conditions under which something becomes valuable.

Customers still have limited attention. They do not suddenly have more patience for another feature, workflow, notification or interface because it was cheaper for us to make one. Every additional thing we introduce still asks them to notice it, understand it, trust it and decide whether it is worth incorporating into their lives or work.

Trust is still slow to earn and quick to lose. Faster production does not make customers more forgiving of unreliable, confusing or intrusive products. If anything, the ability to produce more makes restraint more important because customers experience the accumulation, not the internal productivity gain that created it.

Ownership also remains stubbornly human. Something still needs to be maintained, monitored, explained, changed and eventually retired. That is true whether it took six months to build or somebody assembled the first useful version over a weekend.

Accountability has not disappeared either. AI can contribute analysis, generate options and increasingly take actions. When a consequential decision creates a consequence, somebody still carries it.

None of this is an argument for moving slowly. It is an argument for being clearer about what speed is meant to achieve.

Re-underwrite the roadmap at new prices

One of the more useful strategic exercises right now is not asking where AI should be added to the roadmap. It is asking whether the assumptions behind the roadmap still hold.

Some opportunities that were previously uneconomic may now deserve another look. A customer problem that could never justify six months of engineering may justify a two-day experiment. A workflow that was too expensive to serve profitably may become viable because the cost of delivering it has changed.

The reverse is also true.

Some things looked strategically valuable partly because they were difficult to build. If everyone can now produce a credible version, the fact that you can build it may no longer say much about whether you should. What once looked like differentiation may be rapidly becoming table stakes.

Some permanent products may also become unnecessary. A lightweight workflow, temporary tool or configurable agent may solve a problem without creating another long-lived product surface that customers and teams have to carry indefinitely.

AI should therefore cause us to revisit the basis on which existing bets were approved. What has become newly viable? What has become easier for everybody else too? What can now be tested rather than debated? What no longer justifies a large commitment? Where has the cost moved away from creation and into ownership, oversight, integration or trust?

Sometimes the strategic consequence of AI will be to build something that was previously impossible.

Sometimes it will be to stop building something that is no longer special.

The capacity decision

There is one final consequence I think is easy to miss.

When AI makes a team faster, the organisation has to decide what the released capacity is for.

The default answer is usually more. More experiments, more features, more parallel work, more things that were previously waiting suddenly becoming possible. Some of that may create genuine value. If every gain in productivity is immediately converted into additional production, though, we have made an important strategic decision without ever really discussing it.

There are other places the capacity could go. Teams could spend more time understanding customers properly. They could remove complexity that has accumulated over years. They could improve the quality of things customers already depend on. They could strengthen the evidence behind a consequential decision before committing to it. They could reduce the number of simultaneous priorities people are carrying.

They could even leave some capacity unfilled.

That may sound almost unnatural in organisations trained to treat utilisation as the goal. Capacity does not have to be consumed simply because it exists.

AI productivity should create permission to remove, not an obligation to produce.

I think this is where the strategic opportunity becomes much bigger than efficiency. The promise of cheaper execution is not simply that organisations can keep themselves permanently full. It is that they have more freedom over where their time, talent and technology go.

AI makes more things possible and more things affordable to try. That is genuinely valuable. Possibility, however, is not strategy.

The advantage will not come from finding ever more things to make possible. It will come from deciding which possibilities deserve commitment, which should remain experiments, which should be allowed to act, and which should never become permanent obligations in the first place.

The cost of execution is changing. The responsibility to choose well is not.

If there is a consequential AI bet on your roadmap right now, our Pressure Test is a good place to think it through.