AI Daily · 2026-09-13

Simon Willison used GPT-6 Astra with ChatGPT Work to generate 5K/10K running routes from his address; the task produced GPX files and an embedded visualization, but the process was largely opaque. After the thread was compacted, he could no longer retrieve the Python code, prompting him to argue that agent systems should preserve pre-compaction text and expose it via tools. Separately, a former Palantir engineer examines the 'forward deployed engineer' role and finds that FDE responsibilities, reporting lines and incentives vary wildly across companies, often inviting the critique that it is just consulting rebranded. Palantir's Project Frontline cultivated around 250 engineers into FDEs, many of whom now lead similar teams at OpenAI, Anthropic, xAI and Anduril. Together the pieces highlight the growing need for visibility in AI-assisted work and clearer boundaries around emerging technical roles.

Ecosystem & Beyond (Products / Agents / Tools / Opinions)

Opinion

⭐⭐ [Opinion] Generating running routes with GPT-6 Astra and ChatGPT Work

Simon Willison asked ChatGPT Work with GPT-6 Astra (Max) to plan 5K and 10K running loops from his home address using OSM data. The task ran for 27 minutes and produced an embedded visualization plus downloadable GPX and GeoJSON files. Asked how it built the route, the model said it used Nominatim to locate the address, Overpass to download local OpenStreetMap roads and trails, and then computed the loops locally. Willison notes he could not see the code it actually ran or the exact details in the ChatGPT UI, and by the time he asked for the Python code the thread had been compacted so it was unavailable; he argues any LLM system using compaction should preserve the pre-compacted text and expose it through agent tool calls. The map display used the visualize skill, which wrote an HTML file embedded directly into the ChatGPT UI, holding the route geometry in a JSON script and rendering it with D3 loaded from an allow-listed CDN.

⭐⭐ [Opinion] The Rise of the Forward Deployed Engineer — and How To Do the Job Right

Drawing on his time at Palantir, Citadel and Kepler, author Vinoo Ganesh argues that the title of forward deployed engineer (FDE) describes jobs with almost nothing in common: at an a16z FDE Fellowship dinner, engineers from Snowflake, Anthropic and various startups each described a different role — sales engineer, quota-carrying rep who can write Python, or something closer to a consultant with a statement of work — with different reporting lines and incentives. That ambiguity, he writes, is why so many discussions of FDEs ask whether the role is just reinventing consulting. The piece then traces Palantir's Project Frontline, describing a mistake he helped make, how it turned him into an FDE, and how the program turned software engineers into forward deployed engineers — roughly 250 people went through it, and many now run FDE teams at OpenAI, Anthropic, xAI and Anduril. He also notes that Palantir was from near the beginning split into Product Development, which built the platform, and Business Development, which held both the technical BD people already called FDEs and non-engineering customer-facing Embedded Analysts.