Intel / post

Four practices for reducing unnecessary code and tests in agent workflows

Source: x.com
Date: Published Sep 2, 2026

My top four slop-fighting best practices for coding agents are below. My overall observations (a) the way coding agents work right now, agents give unit tests too much authority, an bad tests are a huge source of slop (2) try to force your coding agent to simplify as much as possible on a goal, especially if it's possible to quantify the simplification (1) @dillon_mulroy's antislop lint, conveniently available as an npm package - a bunch of oxlint rules designed to reject slop (2) the "useless test deleter" from @_lopopolo- asks your agent to check for tests that are useless, and then refactor your code to stop being needlessly complex to conform to said useless tests (3) Tell your AI "delete as much as possible" before finishing a change h/t @Secondmindsys (4) adversarial goal on a slop metric such as LOC, halstead complexity, etc h/t @josephdviviano

Sep 2, 2026 View on X

Context

Henry Dowling describes practices for reducing unnecessary code, what they call slop, that coding agents introduce. Dowling's central claim: agents currently give unit tests too much authority, so a bad or outdated test becomes a source of bloat, because the writes around the test rather than questioning whether the test itself still makes sense.

The two practices Dowling says they use directly are periodically asking an agent to review whether existing tests are still useful and to simplify code that only exists to satisfy an unnecessary one, and telling an agent to delete as much of what it just added as possible before calling a change finished. Dowling also credits other developers with adversarial approaches that score a change against a raw complexity or line-count metric, though they do not give enough detail about those specific tools to verify or link them independently here.

Terms in this piece · Glossary

AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.

Read the source x.com