Local coding agents are good now, but only if you babysit them

In three linesLocal coding agents are useful for small tasks (fixes, repo reading, file changes) but require constant supervision. User describes iterative workflow: task → tests → check diffs → fix issues. Without oversight, agents produce broken code or drift from objectives.

Summary generated by Claude — human-verified