AI Coding Agents Changed How I Work, Not What I Build

Coding agents have moved from autocomplete to something closer to a junior teammate. You describe a task, the agent reads the repo, edits files, runs tests, and comes back with a diff.

What actually got faster

The boring middle of software: renaming things across a codebase, writing tests for existing behavior, wiring up a new endpoint that looks like the four before it. These tasks were never hard, just long.

What did not get faster

Deciding what to build. Understanding a fuzzy requirement. Noticing that the feature everyone asked for will make the data model worse. An agent will happily implement a bad idea very quickly.

Habits that help

• Give the agent a clear, checkable goal, such as "make this failing test pass".
• Review the diff like you would a colleague's pull request. Do not skim it.
• Keep a short project notes file describing conventions, so you stop repeating yourself.
• Commit often, so a bad run is one revert away.

The takeaway

The skill that matters more now is judgment: knowing what good looks like and spotting when the output is subtly wrong. Typing speed was never the bottleneck, and now it is even less of one.