I'm a data scientist by training, so I'm naturally suspicious of AI hype. But a few tools have quietly become permanent fixtures in how I work — and a few others I dropped fast.
AI as a thinking partner, not an oracle
The biggest unlock wasn't generating code — it was rubber-ducking architecture decisions out loud and getting a fast, structured second opinion. I still own the decision; I just reach it quicker.
What actually stuck
- Codegen for boilerplate — tests, types, glue code. The boring 80%.
- Inline review — catching the dumb bug before it reaches CI.
- Doc search — answering "how does this codebase do X" in seconds.
What I dropped
Anything that asked me to trust output I couldn't verify. In production systems, an unexplainable answer is a liability, not a feature.
Used well, these tools don't replace judgement — they remove the friction between having judgement and acting on it.
