Leonid M's articles
- 13 days agoAI Coding
AI Agent Skills Promise 65% Savings. A Benchmark Found 8.5%.
JetBrains tested Caveman on 82 SkillsBench tasks. Token savings fell short, quality stayed flat, and cost variance erased the gain. What actually moves agent work forward.
- 14 days agoAI Coding
Why I Dropped the Big Prompt and Went Skills-First
A big system prompt overloads AI agents instead of making them reliable. Cairn's Skills-First approach loads only the skill needed now and keeps deterministic work in code.
- 19 days agoAI Coding
From Eval to Your Own QA Agent: Why Generating Code Is No Longer Enough
After eval as a truth mechanism, building Cairn: why AI code generation needs an autonomous verification layer, not just more tests.
- about 1 month agoAI Coding
Eval is not a grade — it's a truth mechanism
How I dived into LangGraph and Langfuse through a single experiment — and what evaluation taught me about testing AI agents.
