Samyak Shah

How is AI changing product management?

Building got cheaper. Deciding what to build did not. Notes from running an agentic development program and prototyping with AI every day.

6 min read

This year I helped start an agentic development program at my last company. Five engineers working alongside AI coding agents reached 90% feature completion in six weeks on a roadmap we had planned for a year. Before that, I built a working prototype of a new product in four weeks, and it was the prototype, not a slide deck, that got the product funded.

That pace changes a product manager's job. Some of the change is obvious. Some of it is not, and the part that isn't obvious matters more.

What does AI change for product managers?

You get validation and alignment faster

A prototype used to take weeks of design and engineering time, so we described ideas in documents and hoped everyone pictured the same thing. Now a PM can put something real in front of people in days.

That works in two directions. Externally, customers react to working software far more honestly than they react to a mockup or a description. Internally, leadership, sales and engineering align faster around something they can click than around a document they each read differently. Both conversations that used to take a quarter can now happen in a couple of weeks.

You can work at a different altitude

Much of a PM's week used to go into producing artifacts: tickets, mockups, specs, status updates. AI takes a large share of that work. The time it frees up should not go into producing more artifacts. It should go up a level, into which problem to solve, for which customer, and why now.

Specs become statements of intent

When agents and engineers can generate implementation quickly, a backlog of user stories that dictates every screen and edge case becomes the bottleneck. The PM's level of work moves up to the epic and the end-to-end workflow. What design and engineering need from a PM is intent: the problem, who has it, the outcome we want, the constraints we can't break, and what good looks like.

Intent is also how you check the work. When code is produced faster than anyone can read it, the question is no longer "was it built?" but "does it do what we meant?"

What hasn't changed?

Good products still come from a PM, a designer and engineers sitting together and arguing about the decisions that matter. Which approach do we take? What do we give up? What do we deliberately not build?

AI makes this conversation more important, not less. It can produce five plausible approaches in an afternoon. Choosing between them, and owning the tradeoffs, is still a human job, and it is the job your team needs you in the room for.

Where should PMs spend the time AI gives back?

If AI gives a PM back a day or two a week, here is where I think that time should go.

Understand the market

Who else is solving this problem, how are buyers' expectations moving, and where is the market heading over the next two to three years? When your team can build almost anything, picking the right thing depends on knowing the landscape better than before.

Understand the customer's workflow as it is today

Not the workflow you wish they had. Sit with the people doing the work and map how it actually happens, including the spreadsheets, workarounds and phone calls. Then ask two questions: what would make them change to the workflow you are proposing, and what would stop them? Most products fail on the second question.

Map the forces acting on your customer

Whether a customer adopts your solution depends less on your product than on what is happening inside and around their organization. Some forces pull them toward change. Others hold them back.

Their internal pullWhat inside their organization wants change: the daily pain, a frustrated team, a champion, a leadership mandate.
Their external pullWhat outside pushes them to change: new regulations, deadlines, competitors, their own customers' expectations.
Their internal frictionWhat inside resists: habits, existing tools, budget cycles, security review, the cost of retraining people.
Their external frictionWhat outside slows them down: procurement rules, contracts with incumbents, compliance, partners they have to integrate with.

A prototype will tell you whether people like an idea. Only this kind of map tells you whether they will actually switch, and what your product, pricing and rollout need to do to tip the balance.

Tie it to the business, now and later

How does solving this pain point serve your company's objectives this year, and how does it set up the next three? A feature that wins a quarter but leads the product somewhere you don't want to go is a cost, not a win.

Know the cost of not solving it

What happens if you don't build this? Customers churn, deals are lost, a competitor fills the gap, or nothing happens at all. That last answer is useful. It is the fastest way to stop a project that would have looked productive and changed nothing.

Keep watching the market and the technology

Both are moving faster than they used to. Keep tracking how your market and the underlying technology are changing, so you can adjust your plan quickly instead of finding out a quarter too late. A roadmap built on last year's assumptions about what AI can do is already out of date.

Why the bar went up

AI made building cheaper and faster. It did not make choosing what to build any easier. If anything, when a team can ship a year of roadmap in six weeks, the cost of building the wrong thing arrives faster too. The PMs who do well in this period will be the ones who use the time AI gives back to understand customers, markets and their business more deeply than before.

Samyak Shah

Samyak Shah is a product leader with 15 years taking products from zero to launch. He led legal AI and public records products at Everlaw, climate and humanitarian aid products at Flexport, carbon accounting at nZero, and SMB fintech at Xero and Capital One. He builds with Claude Code and Cursor, and writes about product management in the age of AI.

Connect on LinkedIn See my work