Don't waste your back pressure ·

Automated feedback systems that help AI agents identify and correct their own mistakes as they progress through tasks. This concept explains why some agent applications succeed at longer-horizon tasks while others struggle - successful ones have structured automated feedback that keeps agents aligned without constant human intervention.

When you manually tell an agent about errors (like missing imports), you're spending YOUR back pressure - your finite time and attention - on trivial corrections. This scales poorly and limits you to simple problems, preventing you from thinking about larger architectural goals.

Giving agents access to build systems and bash commands creates automatic feedback loops. The agent can run builds, read error messages, and self-correct without requiring human involvement for syntactic or compilation errors.

Languages with expressive type systems (Rust, Elm, Python with type hints) are growing in popularity partly because they create automated constraints and feedback. They describe contracts, prevent invalid states, and identify edge cases - all forms of back pressure that guide agent changes.