Logan Kilpatrick

Confidence 0.70 · 1 source · last confirmed 2026-06-15

Product lead for Google AI Studio and the Gemini API at Google DeepMind; previously led developer relations at OpenAI. Within this wiki he is the named source for the “model eats the harness” thesis and the jagged-superintelligence framing, via the Sequoia “Why the Model Eats the Harness” interview.

What he is the wiki’s source for

• “The model eats the harness.” The most-cited claim from his Sequoia interview: scaffolding runs ahead of the model, then the model digests it and the alpha moves elsewhere — “the harness, in the way we think of it today, perhaps won’t be true in 12 months.” A frontier-lab counterpoint to the wiki’s agent-harness corpus (see that page’s Debates and supersession).
• Antigravity as Google’s through line. Describes Antigravity — born from the Windsurf team — as the single agent harness now powering search, the Gemini app, Cloud, and AI Studio.
• Jagged / narrow superintelligence. Coding already “feels like narrow superintelligence”; verifiable verticals (math, finance, science) will cross the bar before AGI.
• Outcomes over eyeballs. Argues Google’s goal is maximizing user outcomes, not eyeball time.

Affiliation

Leads AI Studio and the Gemini API at Google DeepMind (a part of Google). The “first-source promotion” here is a judgment call — he is the substantive interview subject and the named originator of a thesis the wiki tracks against its central agent-harness concept; the dangling-author audit (scripts/lint-dangling-authors.mjs) is unaffected (it tracks only author: frontmatter on ≥2 sources).