This is not career advice. It is a way to describe what you actually do, for when the title on your contract stopped describing it.
Read this bit first
The five roles are Boris Cherny’s, not mine. He runs Claude Code at Anthropic and posted them on 28 June 2026. The definitions in the next section are his, tightened for reading. Everything after that, the questions, the pairs, the stage map, is my extension of his framework. He has never seen it and has not endorsed it.
These are not job titles and Anthropic has not renamed anything. Cherny’s actual point is that the roles cut across function. In his words, some designers match category 1, some 2, some 3, and the same holds for engineers, PMs and data scientists.
You will not be one of them cleanly. He says many people span 2 roles and sometimes 3. If you take a single clean answer away from this, you have taken the least useful thing in it.
The original post: x.com/bcherny/status/2071379474277613732
The five, in plain English
1. Prototyper
What you do. Produce brand new ideas at volume. Most of them never ship, and that is the job working correctly rather than failing.
How you know it is you. A blank page is the easiest part of your week. Your hardest part is the last 20%, and there is a graveyard of half-finished things behind you that you are not embarrassed about.
Where you are dangerous. You start the sixth thing before the first one is real. Teams that are all Prototyper ship nothing and feel busy the entire time.
What you are worth. You are the only one who finds the thing nobody asked for. Every other role on this list needs something to exist before they can act.
2. Builder
What you do. Take a rough idea or a prototype and turn it into a real production version quickly.
How you know it is you. You are happiest with a clear target and a half-working thing to aim at. You care about it holding up under real use, not about whether it was your idea.
Where you are dangerous. You will build the wrong thing beautifully if nobody checks the direction. Speed makes that worse, not better.
What you are worth. You are the conversion rate between ideas and reality. Without you a Prototyper is a hobbyist.
3. Sweeper
What you do. Clean up the interface, simplify the system, remove the features nobody uses, make it faster.
How you know it is you. Mess bothers you more than absence does. Deleting something feels as good to you as shipping something, and you have argued to remove a feature someone was proud of.
Where you are dangerous. You can cut something that had a small, quiet, important set of users. Removal needs evidence in a way that addition does not.
What you are worth. You are the only role that reduces the surface area. Every other role adds to it. Without you a product accumulates until nobody can change it safely.
4. Grower
What you do. Take something that already works and push it until it actually fits the market.
How you know it is you. You would rather run the tenth iteration of one thing than the first of ten. You know your numbers without opening a dashboard, and a plateau irritates you more than a failure does.
Where you are dangerous. You will optimise something that should have been killed. Local improvement can hide the fact that the whole direction is wrong.
What you are worth. You are the difference between a product that exists and a product people choose. Almost nothing works on the version that first shipped.
5. Maintainer
What you do. Own a mature system and keep it secure, reliable and fast as it scales.
How you know it is you. You think in failure modes. You are calm when things break because you already imagined this one. Your best weeks look, from the outside, like nothing happened.
Where you are dangerous. You can make change so expensive that the product stops moving. Stability has a cost and it is usually paid in speed.
What you are worth. You are the reason the thing is still there in 3 years. This is the least visible role on the list and the most expensive one to be missing.
Work out which one you are
Copy the prompt below and paste it into Claude or ChatGPT. It asks you 6 questions one at a time, then tells you your primary role, the one you span into, and the blind spot that pairing creates. Takes about 5 minutes.
Answer with what you actually did, not what you meant to do. A hedged answer gets you a mushy result, and the whole point of doing this in a chat rather than on a page is that it can push back on you.
You are going to work out which of five working roles best describes me.
THE FIVE ROLES
1. Prototyper. Generates brand new ideas at volume. Most never ship, and that is the role working correctly.
2. Builder. Takes a rough idea or prototype and turns it into a real production version, fast.
3. Sweeper. Simplifies, cleans up, removes the features nobody uses, makes things faster.
4. Grower. Takes something that already works and iterates on it until it fits the market.
5. Maintainer. Owns a mature system and keeps it secure, reliable and fast as it scales.
These are not job titles. A designer, an engineer, a PM and a data scientist can each be any of the five. Do not use my job title as evidence.
HOW TO RUN THIS
Ask me the questions below ONE AT A TIME. Wait for my answer before asking the next one. Do not tell me which answers map to which role, and do not give any interim assessment until all 6 are answered.
If an answer is vague, hedged, or describes an intention rather than a thing that happened, ask one short follow-up before moving on.
THE QUESTIONS
1. Describe the last thing you worked on that you were genuinely proud of. What state was it in when you picked it up, and what state was it in when you put it down?
2. Someone hands you a product that already works fine. What is the first thing you find yourself wanting to do to it?
3. Think about your last 3 months of work. What did it actually produce? Be concrete.
4. Which of these makes you volunteer before anyone asks: a blank page, a half-built thing that stalled, a mess everyone has stopped noticing, a number that has flattened, or something on fire at 2am?
5. What kind of work do you quietly avoid, or hand to someone else whenever you can?
6. What do people repeatedly praise you for, and what do people repeatedly get frustrated with you about?
THEN GIVE ME
- My primary role, with 2 specific things I said as the evidence.
- The secondary role I span into, with evidence. Almost everyone spans 2.
- The blind spot that pairing creates, stated bluntly.
- One thing I should deliberately get better at, and one thing I should stop trying to be good at.
- What stage of team I am suited to right now, and what stage I would struggle in.
Be direct. If my answers point somewhere I clearly do not want to hear, say it anyway. Do not tell me I am a bit of all five, because that is the useless answer.
Start with question 1 now.If you want to run it across a team, get each person to run it themselves rather than answering on their behalf. Question 5 and question 6 are the two that actually separate the roles, and they are the two people answer differently about themselves than you would answer about them.
The pair you probably span, and the blind spot it hides
Cherny says many people span 2 roles. The pairs are not random. These are the 5 that show up most, what each one is unusually good at, and the thing it cannot see.
Prototyper and Builder. The founder pair. You can go from nothing to something real without waiting for anyone. Your blind spot is subtraction. You will keep adding, because both halves of you are additive roles, and in about 18 months the product will be unmaintainable and you will blame the code.
Builder and Sweeper. The craft pair. What you ship is clean and stays clean, and other people’s code gets better after you touch it. Your blind spot is direction. You will execute and refine your way through a whole year of building something the market did not want.
Sweeper and Grower. The product pair. You improve things by removing friction and you can tell the difference between a feature and a fix. Your blind spot is the new. You will squeeze the current thing for everything it has and never start the next one.
Grower and Maintainer. The scale pair. You are who a company needs at 50 people and rarely who it needs at 5. Your blind spot is urgency. Careful iteration and defended stability both slow things down, and you will not feel it happening.
Prototyper and Grower. The odd pair, and the one that surprises people. You skip the middle. You can find the idea and you can tune it once it works, but the unglamorous conversion between the two is the part you avoid. Your blind spot is that you will keep hiring around it instead of admitting it.
If the pair fits and the single role did not, trust the pair.
Which role a team actually needs, by stage
This is the part that makes the framework worth something. A role that is critical at one stage is dead weight at another, and hiring the wrong one is the most common mistake here.
No product yet. Prototyper, then Builder. Nothing else. A Sweeper has nothing to clean, a Grower has no numbers, and a Maintainer is guarding an empty building.
First real users. Builder, heavily. One Prototyper is useful and two is a problem. This is where teams break, because the thing that got them here is the thing they now have to stop doing.
It works but growth is flat. Grower, and a Sweeper alongside. Almost always the answer is removal before addition, and almost always the team wants to add.
Growing but creaking. Maintainer, urgently, and it will already be later than it should have been. This is the hire that is obvious in hindsight and invisible in advance.
Mature and slow. Prototyper, brought back deliberately and given room. The risk at this stage is not instability, it is irrelevance.
Most teams over-index on Builder at every stage, because Builder is the one that looks like work.
What to do with this
Name your role out loud, once, to someone who works with you. The value is not the label. It is that they now know what to hand you and what to hand somebody else.
Get deliberately better at one, not passable at five. The whole argument here is that depth in a role beats a title. Spreading yourself across all five is just the old game with new words.
Check your role against your team’s stage. If you are a Prototyper on a team that needs a Maintainer, you are not underperforming. You are misplaced, and those are different problems with different fixes.
Know which role you are not. The blind spots above are more actionable than the strengths. Nobody ever got caught out by the thing they were good at.
If you disagree with the stage map, or you think there is a sixth role, DM me. Both are welcome.