ML//agent//active perception

A programmer does not read an entire repository before changing one function. They search, inspect the likely file, follow a dependency, run a test, and use the failure to decide what to inspect next. That is active perception: acting in order to choose the next observation.

A programmer does not read an entire repository before changing one function. They search, inspect the likely file, follow a dependency, run a test, and use the failure to decide what to inspect next. That is active perception: acting in order to choose the next observation.

For an agent, search, read, and inspect are epistemic actions. They reduce uncertainty without directly completing the task. Editing and deploying are world actions. Running tests sits between both categories.

This resembles a partially observable decision process: the real project state is hidden, tools produce observations, and the policy chooses the next acquisition or intervention.

Dumping the whole repository into the context window removes this useful control loop and can bury the evidence that matters.