Home / Blog / Explainers

Explainers

Impact analysis for AI agents: know the blast radius before you edit

Impact analysis walks the call graph from a symbol to everything that depends on it — callers, files, tests. Giving that list to an agent before it edits turns "fix the test that broke" loops into one correct change.

Published September 22, 2026 · 4 min read

Agents break what they didn't read

An agent changes a function signature, updates the two call sites it saw, and misses the third in a module it never opened. The test run fails, the agent reads the failure, opens more files, patches again. Each loop costs thousands of tokens.

How impact works in Finder

Finder's impact tool takes a known symbol and traverses the call graph upstream. It returns the affected files, symbols and tests, along with the traversal evidence for each — so the agent can see why a file is in the list.

• Local and free. Pure graph analysis: no embedding call, no credits.
• No guessing. An ambiguous name returns candidates instead of picking one.
• Honest about gaps. If the graph is still building it says warming — "not ready" is never reported as "no impact".

Pair it with context

context shows one symbol's definition, callers, callees, snippets and tests. Together: context to understand, impact to scope, then edit.

In code review

The opt-in detect-changes tool maps a git diff to changed symbols, their impact and the tests to run — a cited starting point for reviewing a branch.

Frequently asked questions