Google DeepMind's Logan Kilpatrick: Why the Model Eats the Harness

Another new idea to push the state of AI architectures forward. Sakana released a model that effectively uses a mixture of models to get work done. You get a single API but then the work gets farmed out the model that best performs the task. “Fugu manages model selection, delegation, verification, and synthesis automatically. It solves tasks directly when that is enough, or coordinates a team of expert models when a problem calls for more. The complexity of a multi-agent system never reaches your code.” This is generally how applied AI products are building their agent harnesses at this point,

I was skeptical about the multi-model routing. Seems my hinch was right. https://t.co/oVUUqBYRyy

I think one underestimated thing when we look back on it was how useful it is to have your own personal brain and company brain in 2026 at the dawn of usable AGI AGI gives you the intelligence You still have to collect your personal context to get the real unlock

I’m talking about context, not prompt

A good rule of thumb for preventing AI slop (in writing, design, etc): Is your input (the context) longer than the output? I've found that for the AI to produce quality results, my input is often 3-5 times the length of the output. If your input is much shorter than the output, it's almost certainly going to produce slop

Agents will use software 100X more than people. When that happens, theres a huge need for guardrails on what the agents are doing so they don’t leak data or change the wrong information, authoritative sources of truth for them to work with, logging and auditing of what they’re doing, the ability to collaborate with people through these systems, and more. A simple query on any given agentic task could pull in more data than a user touches in a month. As a result, there are lots of categories of software that when it goes headless that the usage and value go up substantially. Agents will end up

Coding agents will squeeze every ounce of IKEA effect out of you, if you let them.

Why HTML turned out to be the foundation for agentic video making from @liu8in: “We’ve been trying to build a video agent. However, we learned the hard way that agents have no visual intelligence. So that’s when we turned to code. HTML is the LLM’s native language. LLMs can express not only information, but also visual aesthetics through HTML, CSS, and JavaScript. Footage, images, assets, SVGs can all sit on top of HTML." 📌 As an example, see how you can make a product video from any website URL here: https://t.co/Q2p8XzCfc6