What you feed AI matters more than how you ask - Into the MoTaverse - Episode 22

Adam Davis is a software engineering manager who spends his spare hours building pixelmesh. It’s a system that turns a room full of phones into a single coordinated light display, inspired by the glowing wristbands at Coldplay concerts.

In this episode he walks through the engineering, the testing, and the hard-won lessons from demoing it live at the Manchester MoTaverse chapter meetup.

The conversation moves quickly from the technical (Morse-code blink detection, region-of-interest camera clipping, eighty-metre range tests from a cul-de-sac at midnight) to the bigger idea: that the prompt engineering craze has distracted people from what actually matters in AI-assisted work. When AI cannot physically see your problem, no amount of clever prompting helps. What helps is observability — structured telemetry, pixel data, debug runs, video replay. Adam explains how treating observability as a day-one practice changed his project, cut costs dramatically, and gave him the data to make AI a genuine debug partner rather than a code generator.