AI Radar

Five months of human–AI collaboration

Gerd Kortuem
TU Delft Faculty of Industrial Design Engineering
Published 8 July 2026

This visualization shows five months of my work with AI agents, between February and July 2026. I set up and worked with roughly 28 agents across three platforms: Claude Code, Cowork, and Codex. The agents ran on frontier models from Anthropic (Opus and Fable) and OpenAI (Codex).

At the centre, the shared memory gathers as the work goes on: the notes, rules, and records the agents wrote and reused, one mark for each write. The points arranged around it are the agents, roughly 28 of them, grouped and coloured by project. Each fine line is an agent reading from or writing to that memory. The sparks are the short-lived worker instances they spawned to carry out pieces of a task. The beam sweeps out from the centre and marks which agent I was interacting with at each hour.

Open the piece in a new tab for a full-viewport view.

How I made it

This visualisation combines data from log files of Claude Code, Cowork, and Codex. I used Claude Fable 5 and Opus 4.8 for data processing and analysing, and for programming the visualisation.