Five patterns, prototyper, builder, sweeper, grower and maintainer, describe what someone does to a product better than a job title does, and the mix a team needs shifts as the product matures.
- Each pattern is different work. Prototypers try many ideas that mostly do not ship, builders make one production-grade, sweepers simplify and remove, growers refine a decided product, and maintainers keep a mature system secure and reliable.
- The mix should change with maturity. A new product leans on prototyping, building and sweeping; a mature one leans on sweeping, growing and maintaining.
- Sweeping is needed at every stage. It is the work of removing things, and it is easy to overlook because its output looks like subtraction.
- Set up AI by pattern, not by title. A prototyper wants loose, fast loops while a maintainer wants permission tiers and an audit trail, and handing either the other's setup slows them down.
- Treat it as a lens, not a study. It is one person's observation of one team, closed with a question about future roles rather than a finding.
As AI takes over more of the typing, the useful way to describe someone stops being their job title and starts being what they are good at doing to a product.
Five patterns keep appearing. Most people are two of them. Which ones your team needs depends entirely on how mature the product is.
The five
Prototyper
Comes up with new ideas and churns out many of them — most of which do not ship, and that is the point rather than a failure.
The output is not the prototype. It is the information about which direction is worth taking seriously. Measuring a prototyper on ship rate destroys the function.
Builder
Turns a prototype into production-grade product or infrastructure. The gap between "it works in a demo" and "it works for strangers at 3am" is the entire job, and it is almost always larger than the prototype took.
Sweeper
Cleans up the interface, simplifies the code and the system, optimises performance — and unships things.
That last verb is the unusual one, and the most valuable. Almost no organisation has anyone whose job is removing what should not have been added. Every accumulated feature is a maintenance cost that nobody chose deliberately.
Grower
Takes something already built and iterates it toward product-market fit. Not new ideas — better versions of a decided one, which is a different temperament from prototyping and frequently mistaken for it.
Maintainer
Owns a mature system and keeps it secure, reliable, fast and efficient as it scales. The work is least visible when done best, which is why it is chronically under-credited and the first thing cut.
The staffing rule
The observation that makes this more than a personality quiz: the mix should change as the product matures.
The sweeper is needed in full at all three stages. The builder appears in all three too, but only as “some” at strong PMF.
Worth sitting with. The sweeper is the only role the rule calls for in full at every stage — and its output looks like subtraction, which is easy to overlook.
Why this matters for how you work with AI
These are not job titles, and the source is explicit that they cut across function — some designers are prototypers, some are builders, and the same is true of engineers and PMs.
The practical consequence: the right AI setup differs by archetype, not by role.
- A prototyper wants loose loops, throwaway scaffolding, and permission to generate ten things badly. Guardrails slow this down for no benefit.
- A builder wants tight briefs and a review order, because the failure mode is shipping something that only worked in the demo.
- A sweeper wants trace reading and a sense of what checking costs — the job is finding what to remove.
- A grower wants measurement and an awareness of sycophancy, because iterating on agreement is iterating on nothing.
- A maintainer wants permission tiers, a security checklist, and an audit trail. Speed is not the constraint.
Handing any of them another's setup makes them slower. That is the useful thing here — not the taxonomy, but the fact that "best practice" is different depending on which of the five you currently are.
Boris Cherny, creator of Claude Code at Anthropic, posted this on X on 28 June 2026. It was picked up widely the following day — Business Insider, Inc., and coverage in at least four languages.x.com/bcherny/status/2071379474277613732 — not read at source: X returns HTTP 402 to automated fetch. Business Insider’s write-up, dated 29 Jun 2026, carries the post — syndicated copy. Non-English pickup checked: Spanish, Ukrainian, Japanese and Korean. Re-checked 11 Sep 2026.
The definitions on this page are his, close to verbatim. Two details worth keeping because they are load-bearing:
- "Many people span across 2 roles, and sometimes 3." Not a sorting hat.
- "These roles are not really tied to job function" — across Anthropic, some designers match category 1, some 2, some 3; the same holds for engineers, PMs and data scientists.
And he closes with a question, not a claim: "Maybe product roles of the future will look more like this, and less like the domain-specific roles of today?"Closing line confirmed in coverage, which renders it without the question mark: "Maybe product roles of the future will look more like this, and less like the domain-specific roles of today." paddo.dev. Re-checked 11 Sep 2026.
So this is one person's observation of one team, offered tentatively — not a study. Its value is that it names something many people recognise and had no words for. Treat it as a lens.
First-hand, as of 23 Sep 2026: one figure not verified — the view count was taken from a screenshot and has not been independently checked. It is omitted from this page rather than repeated.