I talked with Peppe Silletti about what happens when writing code gets cheap and the expensive part is still deciding what to build.
Peppe is an independent product engineer and host of The Product Engineers Podcast. He has spent the last decade in startups, scale-ups, and consulting, and he is still treating the product-engineer role as an exploration rather than a settled job title. That is useful. A lot of us are in the same spot.
The risk did not get smaller. Shipping got cheaper.
Peppe's startup vs scale-up split is simple. Early on you explore. You try paths. You watch the money. After product-market fit, the work tilts toward bottlenecks: the database is slow, the servers cannot take the requests, the war room gets louder. The decisions you made while exploring become the constraints you live with later.
AI changes the first half more than the second. The old lean-startup move was to test assumptions with the cheapest possible experiment. That principle still holds. What changed is the floor. You can ship more than a classic MVP, then remove what nobody uses - the Instagram story of killing everything but photos. You can also ship so much that the data gets noisy and you cannot tell what worked.
The line that stuck with me: AI is a tool, not a silver bullet. If you skip customer discovery, you can now hit the wall faster, with more features, while the model tells you that you are doing great.
A PM holds the map. A product engineer owns a slice.
Peppe pointed at PostHog as a useful starting point, not a religion. They ran for a couple of years with product engineers and no product managers. When they scaled, they added PMs to hold the bigger picture. Product engineers still made the product calls.
His working model is the same shape. The PM is a compass: market, pricing, competitors, whether the whole product is coherent. The product engineer (or a small pair of them) takes one outcome - increase activation by 50% this quarter - and owns it end to end: talk to the users of that problem, prototype, ship, measure.
That is not "now you are management." It is the opposite of taking tickets over the wall. It is also why backend engineers are not off the hook. Your users might be other developers calling your API. Ergonomics is UX. So is making the one endpoint that matters to the business actually fast.
The skills that stay expensive
If agents keep eating implementation, Peppe would start with first principles. Learn to interview without leaping to a solution. Study the basics of UX - he is reading Laws of UX - even if you never design a pixel. Understand your layer and the neighboring layers.
He has been hosting a product-engineering podcast longer than I have. The surprise for him: a lot of teams have been doing this for years under names like extreme programming. They sat with customers. They just did not call themselves product engineers.
The homework
Peppe made it social on purpose.
Ask your PM if you can shadow customer interviews. Watch session replays with your team for a half hour a week. Then get into discovery before the roadmap is already set, so you can say "we could do this simpler and have impact sooner."
That is how you stop being a ticket-taker without pretending AI will do the product thinking for you.
