StepOne
Most CTOs talk about AI productivity gains in the abstract. Mohit Aron, Founder of Cohesity and Nutanix, has a concrete system — one he built by taking his pre-AI coding methodology and mapping AI directly onto every phase of it. He still writes unit tests, runs two coding agents in parallel for different tasks, and has a firm view on why hiring for AI fluency before engineering fundamentals is a trap that will cost you. In this conversation with Ashish, Mohit shares his full AI-assisted development methodology — natural language spec first, then API boundaries, intermediate state, skeleton functions, implementation, and only then unit tests — and why skipping that order is exactly how teams end up owning code they don't understand. He also pushes back on blanket productivity claims: 20x gains are real for unit tests and pattern-heavy work, but design-heavy work with humans in the loop will see far less. And on the prediction that 90% of engineering jobs disappear in 18 months, he's not buying it — his company is still actively hiring, and good engineers remain hard to find. Topics discussed: * Early-stage CTO as player-coach: when to build alongside the team vs. step back * The 5-step AI coding methodology: spec, API, intermediate state, skeleton functions, implementation * Why open-ended AI prompting produces bad output and who is accountable when it does * Applying a looser review standard to AI-written unit tests vs. production code and why * Interviewing engineers without AI to test fundamentals — fluency is a bonus, not a baseline * Why high token usage means nothing if the output quality is poor * Running multiple coding agents in parallel and switching between them by task type * Why AI as a pattern matcher still requires strong engineers to catch what it gets wrong
6 episodes
Comments
0Be the first to comment
Sign up now and become a member of the StepOne community!