Why (Senior) Engineers Struggle to Build AI Agents
summary
Claim: senior engineers fail because agents lack years of implicit context — an agent only sees the deleteItem endpoint schema and docstring, not the lived assumptions behind it. Four concrete shifts follow: text replaces structured state, errors are inputs, evals replace unit tests, and build to delete. Long-running runs (5–15 minutes) make restarts costly, tools must be self-documenting, recovery paths replace retries, and success rates (how often something works) matter more than deterministic correctness.
AI-written summary · about
chapters (7)
- 0:00Agent-building challenges
- 0:50Text as state
- 2:14Deterministic to dynamic control
- 4:01Errors as inputs
- 6:28Unit tests to evaluations
- 7:53APIs for evolving agents
- 9:21Reliable agent principles
related talks
20mThe Three Ways AI Agents Fail. Here's How to Fix Yours.
01h 00mKenote: AI Didn’t Replace You. It Promoted You!
41mThe Ironies of AI²
27mEngineering in the Age of AI: Making AI Work for Us, Not Replace Us
09mLessons Learned in AI Engineering
39mFrom Demo to Production: Why Agentic AI Systems Fail - and How to Fix Them
36mHow to Get AI to Stop Being Weird and Actually Be Useful
14m15 Months of building AI agents in 15 minutes