AI didn't make me faster, it made me better. Here's the data.
Two years into using AI tooling daily across real client work and personal projects. The thing I keep noticing is that the speed-up is the wrong story. The artifact at the end is the actual story.
Last week three of us rearchitected a platform UI in thirty minutes on a video call. The interesting part isn't the thirty minutes. It's that the rearchitecture was documented, the test plan was updated, and the next person to open the project the following morning could read both the new design and the rejected one with the reasoning we'd talked through. None of that fits in the LinkedIn version of "I built [thing] in 30 min with AI."
Two real builds I've been working through, both with Cursor + Claude doing most of the heavy lifting:
- A four-week team build of a Salesforce monitoring platform. Conventionally a 12-18 month, multi-developer build. Real product, real clients ahead of it.
- A two-month solo build of an internal PM/time-tracking platform replacing two seasoned vendors. ~91 seeded users, six teams, full resource planning + reports + in-app AI assistant.
Both came out better than the same effort would have produced two years ago. More thoughtful. Better documented. Better tested. Decisions captured as I made them, including the ones I later reversed. The speed is downstream of the discipline, not upstream.
METR ran an RCT and found experienced devs were 19% slower with AI but believed they were 20% faster. Apollo ran the same question across 250+ engineers and landed at 1.15x with the takeaway that the differentiator was context infrastructure, not the model.
Wrote up the longer version with the case studies, the cost math, and what I think people should actually do with this: https://ianbezanson.ca/posts/stop-measuring-ai-by-the-stopwatch/