Oct 16, 2026 · 1h 0m
AI writes production code but still struggles to run it
AI Can Write the Code. Can It Run the System? | Grepr CEO Jad Naous
The episode examines whether AI can move from accelerating software development to safely operating the complex, failure-prone systems that software creates.
- 1Experienced engineers gain the most from coding AI because they can decompose complexity, direct agents, and judge production tradeoffs.
- 2AI operations becomes more plausible when systems expose clear signals and teams can encode investigations as repeatable procedures.
- 3Coding assistance has an obvious economic payoff, while AI operations remains harder to define, automate, and sell to customers.
Don't miss
Jad Naous explains how he largely built an auto-promotion system for data lakes with AI assistance, compressing work that once required months and a team into weeks.
The brief
Martin Casado and Grepr CEO Jad Naous frame a sharp divide: AI can generate production code, but operating the distributed systems behind it demands far more than plausible syntax.
Naous says experienced engineers get disproportionate leverage from coding agents because they can split complex work, coordinate multiple sessions, and evaluate whether the result is production-ready.
His AI-assisted workflow uses Claude Code, WSL, command-line agents, design documents, reviews, tests, and pull requests to let a small team ship work far faster.
The harder frontier is operations: detecting novel patterns in huge log streams, investigating rare failures, weighing security risks, and deciding when an intervention is safe.
The episode’s conclusion is conditional rather than triumphant: AI operations may work where systems expose strong signals and repeatable procedures, but its value remains harder to explain than coding productivity.