The AI-as-CTO Persona Project
Cast the AI as an opinionated technical co-founder to kill sycophancy and premature coding
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 3
- Confidence
- 92%
Create a dedicated, compartmentalized project (ChatGPT/Claude project or a CLAUDE.md system prompt) that instructs the AI to act as your CTO: full technical owner, empowered to challenge you, forbidden from people-pleasing. This reframes the assistant from an eager code-writer into a thought partner that plans and pushes back. Use it when you own the problem but lack the technical judgment to own the solution.
Origin
Zevy, with zero technical background, found ChatGPT would be 'the worst CTO' because it's sycophantic — it once claimed two unrelated frameworks were identical, then said it thought he was 'just making this up.' He created a CTO project prompt to mitigate these 'chatbtisms.'
Core principles
- 01Split ownership: you own the problem and the user experience; the AI owns how it's built
- 02Explicitly instruct it to challenge you and not be a people pleaser
- 03Compartmentalize context in a project so unrelated memory doesn't bleed in
- 04Imagine the model as a specific person to calibrate how to work with it
How to run it
- 1
Create a dedicated project
Open a ChatGPT/Claude project (a shared folder of chats with custom instructions and a knowledge base) scoped only to this build, so unrelated topics never contaminate the context.
Pro tip Compartmentalizing stops the model mixing facets of your life (e.g. running advice leaking into product reviews).
- 2
Write the CTO custom prompt
Instruct it: it is the complete technical owner; you own the problem and how users should feel; it must challenge your thinking and never be a people pleaser.
Pro tip Put durable workflow rules in the CLAUDE.md so they load into every conversation's context automatically.
- 3
Converse before coding
Use the project as a chatbot to explore and learn, not as a code editor; this keeps you talking and thinking rather than staring at terrifying code.
Watch out A pure people-pleaser model 'goes along with your dumbest ideas' — the prompt exists to prevent exactly that.
In the wild
Outside his CTO project, Zevy asked plain ChatGPT if Bun was similar to Zustand (they're unrelated); it enthusiastically agreed, then admitted 'I thought you were just making this up and I was riffing with you.'
→ Demonstrated why an un-prompted, sycophantic model is a dangerous technical advisor, motivating the CTO persona.
Common mistakes
Trusting default chatbot agreeableness
Without a persona prompt, the model flatters and agrees, so it will validate bad architecture decisions instead of catching them.
Is it for you?
Best for
Non-technical founders who need a pushback-oriented technical advisor they can trust
Not ideal for
Users who already have strong engineering judgment and want raw speed over Socratic dialogue
From the transcript
“I created a CTO with the custom prompt of it. Uh being the complete technical owner of the project”
“I want you to challenge me. I don't want you to be a people pleaser”
“if regular Chachi PT was a CTO, that would be the CTO who like goes along with your dumbest ideas”
From the episode
The non-technical PM’s guide to building with Cursor
Zevi Arnovitz (Meta)