4. How product changed since AI

§The first three chapters described product as a responsibility held by one person. This chapter is about why that description, which used to be a startup exception, is becoming the general case. Between 2023 and 2026 the job titles that carried product in larger companies were renamed, merged or hollowed out. Understanding how tells a founder what the job actually is once the tools stop hiding it.

§ 4.1The roles, before#

§For roughly twenty years, a product team in a software company was a triad. A product manager decided what to build and why, and wrote it down. A designer decided how it should look and behave. Engineers built it. Around them, in larger organizations, sat a product owner who ran the backlog for a delivery team, and above them a product leader who owned the strategy.

§The division existed because each step was expensive and needed a specialist. Writing a good requirement took judgment; drawing a good interface took craft; building took code. No one person could do all three at production quality, so the roles were separate, and the coordination between them was the product manager's day.

§ 4.2What is happening to each#

§The product manager is splitting. Marty Cagan, who wrote the book most PMs learned from, published two years into generative AI that the roles built on gathering requirements, writing tickets and managing a backlog were the most exposed in the industry, and that the product owner role in particular was, in his words, a sitting duck. At the same time the market for product people who own outcomes has not shrunk. Lenny Rachitsky's job-market analysis for early 2026 counted product openings at a three-year high, with two categories growing fastest: growth product managers, and a new title, the AI product manager, which by his count had become a hiring category of its own. The middle of the old role, the coordination, is going. The two ends, deciding what matters and owning the growth of it, are growing.

§The product engineer is arriving. A growing number of companies now hire engineers who carry product judgment, decide within boundaries without waiting for a PM, and understand the customer's workflow well enough to change it. A CIO piece from mid-2026 describes the shift from a CTO's seat: as implementation gets faster, the bottleneck is no longer coding but the organizational friction around it, so the answer is fewer handoffs and engineers who decide. The title varies, product engineer, full-stack builder, but the shape is the same: one person, the whole loop from problem to shipped change.

§The designer is becoming a design engineer. The title existed at companies like Vercel and Linear before the models arrived, and the models made it common. A design engineer designs and ships; the handoff from a mockup to a developer, which was the most expensive seam in the old triad, disappears when the same person can produce production-ready interface from intent. Writers in the design community have described the pressure honestly: production roles and fast-turnaround freelance work are shrinking, and what remains is direction and taste.

§A new role appeared at the customer's desk. The forward deployed engineer was a Palantir invention from the early 2010s, an engineer who embeds with one customer and builds in their systems rather than shipping one product to many. In 2025 it became, in Gergely Orosz's reporting, the hottest job in tech: OpenAI built a team of them across eight cities, Ramp and others followed, and the reason is the one this whitepaper keeps returning to. When building is cheap, the scarce skill is understanding what a real person, in a real system, actually needs. The FDE is discovery and delivery in one seat.

§ 4.3The pattern under the titles#

§Read the four together and the pattern is not that AI replaced product. It is that AI removed the cost that justified splitting product into roles. When a requirement can be drafted in a minute, a screen generated in a minute, and a working prototype in an afternoon, the coordination between three specialists costs more than it saves. The work flows to whoever can hold the whole loop, and the market renames that person.

§That is why the large-company product literature, written for the triad, reads strangely at a startup now. It is describing coordination that a founder never had to do and that larger companies are learning to do without.

§ 4.4The same collapse, at a startup#

§Startups never had the triad. A founding team of two to five held product, design and engineering between them because there was nobody else. What changed is that the tools now assume it.

§In early 2025 Andrej Karpathy gave the practice a name, vibe coding, and the fastest-growing tools of that year were the ones that let a founder describe a product and have it exist. Carta's data on new companies shows the consequence: the share of startups formed with a single founder rose from under a quarter in 2019 to more than a third in the first half of 2025. Base44, a tool for building software from a description, was itself built largely by one founder and acquired by Wix within about six months of launch. Anthropic's Dario Amodei has publicly put better-than-even odds on the first one-person billion-dollar company appearing by 2026. Whether or not that arrives on schedule, the direction is not in doubt.

§For a team of two to five, this means the roles in chapter 2 are not a division of labour so much as a division of accountability. Everyone will build, everyone will talk to customers, everyone will touch the interface. One person still decides.

§For a solo founder, the full product stack is not a metaphor. The same person holds the vision, decides the scope, draws the structure, writes the prompt that produces the interface, ships it, watches a stranger use it, and reads the cohort. There is no one to hand off to, and no coordination cost, which is the advantage. The risk is the one chapter 10 describes: owning everything and deciding nothing, now with twenty prototypes instead of one backlog.

§ 4.5What to take from this#

§Three things. The job described in this whitepaper is the job the market is now hiring for under new names, which means a founder who learns it is learning something transferable. The large-company product canon should be read for its ideas about customers and risk, not its ideas about coordination. And the collapse of roles raises the stakes on the one thing that did not collapse: someone, by name, deciding what gets built from evidence.

§Part II is about the three disciplines that used to be roles and are now, at a startup, part of the owner's stack: user experience, user interface, and the working relationship with engineering.