Prototyper, Builder, Grower, Sweeper, Maintainer: The Roles I See Emerging on AI-Native Teams
Boris Cherny, the creator of Claude Code, recently described five roles forming inside Anthropic's own engineering org once AI coding tools let everyone write code. After two decades scaling tech teams in regulated industries, I recognize the shift, and I want to share what stands out to me, and where it still gets harder.

Date
I came across a LinkedIn post quoting Boris Cherny, the creator of Claude Code, from his appearance on Bloomberg's Odd Lots. I hadn't heard the full episode, just the excerpt, but one detail stopped me mid-scroll.
On his own team at Anthropic, everyone writes code now, designers, product managers and engineering managers included. Not because they became engineers, but because it got easy enough that asking someone else to move a button by a pixel stopped making sense.
For anyone who wants the full context, the episode is here: Odd Lots: The Creator of Claude Code on The Hottest Piece of Software in the World (Bloomberg, 20.07.2026, with Tracy Alloway and Joe Weisenthal).
Five roles, not five job titles
Cherny laid this out first in a post on X at the end of June 2026, then went deeper on Odd Lots. Once code stops being the bottleneck, the old split of engineering vs. design vs. product vs. research stops describing what people actually do. What replaces it, he says, are five archetypes tied to a stage in a product's life, not to a title:
- Prototyper - comes up with brand new ideas and iterates fast. Most of it never ships.
- Builder - turns a validated idea into a production-grade product and gets it to market.
- Grower (he also says "scaler") - takes a product with product-market fit and pushes it further, 10x, sometimes 100x.
- Sweeper - cleans up the UI, the code, the system, removes what shouldn't have shipped.
- Maintainer - owns a mature system and keeps it secure, reliable and fast once it's at scale.
He was upfront that he isn't sure about the names yet. "Sweeper or a janitor, I don't know if there's a better name for it" was as far as he got.
Where this matches something I've been working through
This landed at a useful moment. With my work on Shipwright, I've been thinking about the same question: what the roles look like once an AI harness in combination with loop engineering absorbs most of the mechanical work between an idea and working software. Cherny's framework is a good look at where a team lands once it's through that transformation, not just starting it.
Two things stand out to me about that end state.
First, there's a lot less translation happening. An idea used to move through several hand-offs, PM to designer to engineer to whoever owned deployment, and each hand-off lost something in the process. Now the person with the idea can often go straight to working software themselves.
Second, and this is the part I keep coming back to: it lets teams be talent-oriented in a way they rarely could before. A builder on a small team used to also be the maintainer, and often the sweeper too, simply because the team was too small to specialize. A designer had to cover research, UI and sometimes front-end code, whether or not that was their strength. Or an organization handed everything downstream to one DevOps team, because that was the only way to cover every stage, not because it matched anyone's actual talent. Not everyone is equally good at every phase of building something. Once the mechanical overhead drops, you can finally staff for what people are good at, rather than for what a role definition forces them to cover.
That's a large part of why I'm building Shipwright: a process that runs alongside AI-assisted development and tracks the stage the work is actually in, Specify, Design, Plan, Develop, Validate, Release, Deploy, so a team can organize around who's good at which stage instead of who happens to hold which title.
We're all still learning this
None of this is settled, including for Cherny. I don't think anyone fully knows what these roles turn into until they're on the other side of the transformation. We'll only ask the right questions once we're partway through it ourselves. But the transformation itself isn't optional anymore. The only real choice left is how deliberately we go through it.
Sources:
- Boris Cherny on X, five archetypes for AI-native teams - X - 28.06.2026 - https://x.com/bcherny/status/2071379474277613732
- Odd Lots: The Creator of Claude Code on The Hottest Piece of Software in the World - Bloomberg - 20.07.2026 - https://www.youtube.com/watch?v=7C_IHWkHKmU
- Claude Code's creator on the end of the software engineer - Platformer (Casey Newton) - 26.05.2026 - https://www.platformer.news/boris-cherny-interview-ai-jobs/
- The Five Roles Replacing "Engineer" and "Designer" on AI-Native Teams - The Augmented Work - June 2026 - https://theaugmented.work/articles/roles-not-titles
