SWE Agent
An agent built to resolve real software issues end to end — read the repo, write the fix, prove it works.
Think of it like
Handing a junior engineer a bug ticket and getting back a tested, ready-to-merge pull request.
Example
Given a GitHub issue, the agent locates the buggy files, edits them, runs the test suite, and produces a passing patch.
How it actually works
SWE agents are the benchmark-driven face of agentic coding, measured on suites like SWE-bench where the task is “given this repo and this issue, produce a patch that passes the hidden tests.” Progress here has been fast because the scoring is objective. The gap to everyday work is that real tickets are underspecified and touch design, review, and taste — things a hidden test can’t capture.
For product teams
The clearest yardstick for “can AI actually do engineering work,” with a real-world asterisk.
For engineers
An agent evaluated on repository-level issue resolution, producing patches validated against held-out tests.
Related
- Agentic Coding — The broader practice it’s the sharpest example of.
- Sandbox Execution — The kind of proof it runs to validate a fix.
- Self-Correction — How it recovers when tests fail.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome