Leonid M's articles
- 7 days agoProgramming
I Put a Decision Model in a Playwright Loop. It Missed 2 of 3 Bugs
A model picks the next probe and judges each result while Playwright drives the page. What a small local run showed about confidence, exploration, and what code should check.
- 9 days agoAI Coding
Most Agent Concepts Are Old Ideas With New Labels
Of the thirty terms on the standard reading list, most are ordinary engineering renamed or product categories in disguise. Six have no prior analogue. Those six are the work.
- 10 days agoAI Coding
A Passing Agent Test Says Nothing About the Path It Took
A refund agent returns the right answer and the ledger is correct. The test is green. Why final-state assertions miss what an agent did, and how to check the path.
- 12 days agoAI Coding
When an Agent Misbehaves, Which Layer Broke?
Harness, loop and graph is one more taxonomy unless it changes what you fix. It does, because each layer has its own symptoms, its own test, and one costly misdiagnosis.
