What to automate with Claude Code

Not everything should be automated, but a surprising amount of developer busywork should. Here are the categories worth handing to Claude Code, and where to start.

A simple test for what to automate

Automate work that is repetitive, well-defined, and verifiable. If you can describe the done state clearly and check the result quickly, it is a good candidate. If the task needs taste, judgment calls, or context only you hold, keep it manual or keep a human in the loop.

Tests and quality checks

This is the safest category because the agent can verify itself.

• Backfill unit tests for untested branches.
• Generate fixtures and edge-case inputs.
• Tighten linter and type errors across a module.

Documentation and summaries

Docs are perpetually stale because writing them is nobody's favorite job. Hand off docstring generation, README updates, architecture summaries, and migration notes. The agent turns code into prose patiently and consistently, and you edit rather than write from scratch.

Maintenance and refactors

Mechanical refactors, dependency upgrades, and codemods are perfect to automate because the change pattern is repeatable and the test suite is the judge. Keep the risky judgment calls, such as which major version to adopt, for yourself; let the agent do the typing.

Research that helps a cause, not just you

One category of automatable work gives back. Tokens for Good automates nonprofit vetting: your agent claims a queued organization, researches its impact against a fixed methodology with citations, and submits a structured report. It is genuinely useful work, not make-work, and it runs on your existing Claude subscription with no separate API cost. Add it via npx tokens-for-good init or the remote MCP.

Let it run while you sleep

The point of automation is that you stop thinking about it. Tokens for Good can run on a schedule via /tfg-schedule on Anthropic cloud, so the research continues with your machine off. Each report is researched twice, validated, consolidated, scored on a fixed rubric, and finalized by a human before publishing to the directory. See how the research works.

Frequently asked questions