Honest Impressions After Using Three Coding Agents for a Month

I got approval at work and alternated between three of them for a month. To get straight to the point, the claim that they "do everything on their own" is still overhyped, but if you direct them well, they're genuinely fast.

• Exploration/Explanation: Overwhelmingly useful for getting them to understand the structure of an unfamiliar codebase. All three were similar here.
• Bulk edits: I'm still uneasy about tasks that touch multiple files. If you ask for too much at once, they lose context midway.
• Test writing: The most satisfying part. However, sometimes instead of writing tests that pass, they write "wrong tests," so I always deliberately create failure cases to verify.

What I felt across all of them is that the review cost doesn't disappear. The faster you get code out, the more time you spend reading it. So these days I run them in small increments and look at the diff myself every time.

One-line summary: It's less about replacing juniors and more about tripling the hands of someone who already works well.

by 디지털노마드44