March 29, 2026 · Matt McKenna
Knowledge Follows Code Paths
Why we built code structure awareness into Surchin's retrieval pipeline, and what it means for teams accumulating developer knowledge at scale.
Read more →Blog
Engineering insights from building Surchin.
March 27, 2026 · Matt McKenna
Instructions Are Rigid, Tool Descriptions Are Flexible
We discovered that instructions and tool descriptions serve fundamentally different cognitive roles for AI agents. Enriching tool descriptions with trigger signals improved deposit compliance to 100% on Opus — while making instructions more flexible caused regression.
Read more →March 25, 2026 · Matt McKenna
Three Ways to Poison Your Agent's Context Window
We audited Surchin's retrieval pipeline and found three failure modes that apply to anyone injecting context into LLM workflows.
Read more →March 19, 2026 · Matt McKenna
How Insight Promotion Works
Surchin uses session diversity — not team size — to decide when an insight is ready to promote. Here's the mechanism and why it works for solo devs and teams alike.
Read more →March 6, 2026 · Matt McKenna
Surchin Cuts Agent Costs 21% With Statistically Significant Results
We ran 162 Opus 4.6 agent sessions across Python and Android codebases. Surchin reduced costs up to 21% and made agent behavior 3x more predictable.
Read more →March 5, 2026 · Matt McKenna
When Your Agent Reviews Your Checklist
An AI agent gave us feedback on our CLAUDE.md instructions. We built a benchmark to test its suggestions. The results surprised us.
Read more →March 4, 2026 · Matt McKenna
How We Got Opus to 98% Tool Compliance With a 9-Line Checklist
What we learned benchmarking 20+ CLAUDE.md instruction variants across Claude Opus, Sonnet, and Haiku.
Read more →