23. AI Coding Workflows

The lesson

Good AI coding starts with a precise task, clear file scope, tests to run, and explicit instructions not to touch unrelated code.

Use AI to explore, plan, implement, test, and review. Do not let it refactor unrelated systems just because it can.

For Rust, ask for compiler-safe changes, ownership reasoning, error handling, and minimal dependencies.

A graduate developer at a software consultancy names the exact folder, the test command and the modules the agent must not modify before requesting a change.

A team lead at a healthcare software company asks the assistant to propose a plan before it writes any code for a scheduling fix.

A junior engineer at an appliance manufacturer limits the agent to one firmware file and lists the unrelated drivers it must leave alone.

A payroll systems analyst states the exact failing test and the required result before asking for a fix to a calculation routine.

A support engineer at a logistics software firm asks the assistant to change only the label formatting function and to run the existing test suite afterwards.

An IT administrator at a secondary school scopes a script change to one timetable file and instructs the agent not to touch the shared login module.

A data engineer at a utility records the working directory, the language version and a sample input before requesting a parser change.

A maintenance developer at an insurance firm asks for a minimal fix to a claims import bug with no refactoring of unrelated files.

A junior tester at an online retail team asks the assistant to add the missing unit tests without altering any production code.

A robotics programmer at an automation firm lists the timing constraints and the files in scope before requesting a small control-loop change.

Check yourself