Simon Willison: GPT-6 Astra built 5K/10K running routes in 27 min, but compaction hid the code
dl_weekly · x · 2026-09-24
Simon Willison had ChatGPT Work (GPT-6 Astra Max) generate 5K and 10K looping running routes from his address using OpenStreetMap data. The agent ran for 27 minutes and produced embedded visualizations plus downloadable GPX/GeoJSON files.
Key takeaways:
- It used Nominatim for geocoding, Overpass to fetch OSM roads/trails, then computed loops locally
- The ChatGPT UI gave no visibility into the Python actually executed — which Willison calls an anti-feature
- He later couldn't retrieve the code because the thread had been compacted
- His prescription: systems using compaction must preserve pre-compaction text and expose it via agent tool calls
- Map display used the visualize skill, generating an HTML file embedded in the UI
A neat long-horizon agent demo that also exposes real transparency and context-compaction pain points.
More from coding & agent
- KAIST's WikiTrace: evidence-linked memory fixes agent "memory-induced degradation" — dair_ai · 2026-11-11
- Why inference speed will be AI products' biggest competitive edge in 1–2 years — itsOmSarraf_ · 2026-11-11
- What happened when I moved vector retrieval into Snowflake SQL instead of orchestrating RAG outside — dima806_dima · 2026-11-11
- OpenAI engineer runs Codex for 26 days straight hunting Atlantis, spawning "vibe archaeology" — Polymarket · 2026-11-11
- He replaced a $5,000/month ghostwriter with 11 Claude Skills and earned $47,000 from one post — aryanXmahajan · 2026-11-11
- Refero Styles ships 2,000+ DESIGN.md files to give Claude Code real design taste — bbssppllvv · 2026-11-11