Plan before you let the agent build a feature

A developer let Claude computer use implement a feature on its own, but the output broke and they abandoned the branch.

Done with Claude computer use· Advanced· Oct 2026

What they did
The developer and Claude first settled the feature scope and a concrete delivery plan. After that, they gave Claude full autonomy to carry out the work in the repo. Claude created 70 files without more guidance or test runs, and the developer checked the result afterward.
How it went
The code did not function. They deleted the branch, reset the session, and went back to using Claude for smaller pair-programming steps.
Tips
If you try this, keep the agent on a short leash: ask for a plan first, then require progress checks before it makes large changes. Always pause for approval before any delete, submit, or payment step.

Try it yourself with Claude computer use

In [repo path], help me implement [feature description]. Start by giving me a short plan and the files you expect to change, then wait for my approval. After I approve, work in small steps, run the relevant tests after each step, and pause before any irreversible action such as deleting files, sending anything, submitting a change, or making a payment. Stop once the feature works and tell me exactly how to verify it with [command].

Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.

Read the original on x.com ↗

Get 3 agent uses like this every morning

With the day's top AI news. Free, and one click to unsubscribe.