Experienced developers predicted AI would make them 24% faster. Afterward, they believed it had made them about 20% faster. When METR measured the work in 2025, they were 19% slower.

I have felt that gap in my own work. This session teaches the four moves I use to close it: spec before prompt, small slices that you run every time, feed the model the real error and not your memory of it, and keep the human twenty percent human. Each move is demonstrated live, not explained on slides.

In workshop format, everyone leaves with a working change in their own project and a one-page card of the flow. It is tool-agnostic, with no installs, and works with Claude Code, Cursor, Copilot, or a browser chat window. The goal is not to go faster. The goal is to still be able to explain every function you shipped.

First delivered as a workshop at OwlHacks '26, Temple University, September 2026. Runs as a 30-45 minute talk or a 60-minute hands-on workshop. Attendees need a laptop with their own project; nothing to install. Target audience: students, early-career developers, and teams adopting AI coding tools.