Subvolts take Google's Logan Kilpatrick sees AI models absorbing agent tools, shifting focus to vertical intelligence. Jun 11, 2026 - Developers
- Learn about the predicted short lifespan of agent harnesses and the shift towards integrated model capabilities.
- Business
- Understand Google's strategy for unified AI models and the future of AI agent development.
A Google leader shares a bold prediction about AI agent harnesses and details Google's unified multimodal model strategy.
Google DeepMind's Logan Kilpatrick: Why the Model Eats the Harness
Audio streams from Sequoia Capital. Your place is remembered on this device.
From Subvolts
Summary and analysis by Subvolts editorial, AI-assisted.
Google DeepMind's Logan Kilpatrick discusses the future of AI agents and models on the Training Data podcast. He predicts agent harnesses have a short lifespan as models will run them natively. Kilpatrick also details Google's Omni model, designed to replace separate text, audio, music, image, and video systems, emphasizing AI as an accelerant for human ambition.
AI agent tools, like scaffolding for AI, will soon be built directly into AI models. This means AI will get much smarter in specific areas like math or science before it becomes generally intelligent, helping people achieve more.
This episode offers a glimpse into Google's AI strategy from a key leader. Kilpatrick's prediction about agent harnesses challenges current startup trends, suggesting a shift towards integrated model capabilities. The discussion on Omni highlights Google's move towards unified multimodal AI. This is valuable for developers building AI applications, business leaders strategizing AI integration, and anyone tracking the evolution of AI from specialized tools to more generalized, powerful systems.
- HostSequoia Capital
- GuestLogan Kilpatrickleads Google AI Studio and the Gemini API, Google
- Guest: Logan Kilpatrick, Google DeepMind
- Episode Title: Google DeepMind's Logan Kilpatrick: Why the Model Eats the Harness
- Show: Training Data (Google)
- Length: 51 minutes
- Agent Harness Shelf Life: Roughly 12 months
- Google's Unified Model: Omni
The entire startup ecosystem is racing to build agent harnesses. Logan Kilpatrick, who leads Google AI Studio and the Gemini API, argues that scramble has a roughly 12-month shelf life. Models will absorb the scaffolding and run it natively, so the edge moves elsewhere. Google's own bet runs in parallel: a single agent harness, born from the Windsurf team and now called Antigravity, has become the connective tissue across search, the Gemini app, Cloud, and AI Studio — the role Gemini-the-model used to play. Logan makes the case that coding already feels like narrow superintelligence, and that "jagged" vertical superintelligence (in math, finance, and science) will arrive well before AGI. He argues Google's real goal is maximizing outcomes for users, not eyeball time. He unpacks Omni, the single model built to replace multiple separate systems Google once trained for text, audio, music, image, and video. His throughline: AI is an accelerant for human ambition, not a substitute for it.
Hosted by Sonya Huang, Sequoia Capital
- agent harnesses
- Logan Kilpatrick
- Scaffolding or tools built to help AI agents perform tasks.
- AGI
- Logan Kilpatrick
- Artificial General Intelligence, AI that can perform any intellectual task a human can.
- jagged vertical superintelligence
- Logan Kilpatrick
- AI that becomes extremely advanced in specific fields like math or science, but not generally intelligent.
- multimodal
- Logan Kilpatrick
- AI that can understand and process different types of information, like text, images, and audio.
1 h 08David Silver and Lasse Espeholt of Ineffable Intelligence: A Superlearner for the Age of Experience
WithDavid Silver, Lasse Espeholt
1 h 041 h 0552 minMaking Cities Awesome: Peregrine’s Nick Noone & Ben Rudolph
WithNick Noone, Ben Rudolph
Same guests, other shows
59 minThe Model Eats the Scaffolding: DeepMind's Logan Kilpatrick & Tulsee Doshi on 3.5 Flash, Omni & More
WithLogan Kilpatrick, Tulsee Doshi
1 h 46Radically Better Reasoning: Elicit's Andreas Stuhlmüller & Jungwon Byun on World Models for Research
WithAndreas Stuhlmüller, Jungwon Byun
48 min
Questions people ask
- What is the expected lifespan of current AI agent harnesses?
- Logan Kilpatrick estimates agent harnesses have a shelf life of about 12 months before models absorb their functionality natively.
- What is Google's Omni model?
- Omni is a single model Google built to replace separate systems previously trained for text, audio, music, image, and video.
Page generated Oct 19, 2026. Summaries are Subvolts' own; the audio, artwork and notes belong to Sequoia Capital.