Jerry Tworek & Core Automation, Auto-research is moving from benchmarks to real-world loops

Auto-research: A conversation with Jerry Tworek (CEO @ Core Automation, fmr VP of Research @ OpenAI)

“There is no way to outsource your understanding.”

That feels like the right lesson for auto-research. Jerry traces the path from code models and HumanEval to reinforcement learning, tool use, test-time compute, and automated AI labs.

His core point: as agents get better at writing code, running experiments, and generating more ideas than humans can manually inspect, the bottleneck shifts from generation to verification.

Code became one of the first serious AI verticals because correctness is hard to fake. Static evals helped for a while, but as Jerry puts it, “the era of evals is done.”

The real test increasingly looks like deployment: live feedback, real workflows, and whether a system keeps improving outside the benchmark. So the question is not just whether agents can do research. It is whether we can build verification loops that make their work trustworthy.

More videos, demos, and interviews from Auto Research Build Day are coming soon on the AGI House platform.

What’s Upcoming: Voice Agent Build Day with Inworld

Voice agents are quickly becoming one of the most important interfaces for AI.

The interesting question is no longer just whether a model can speak naturally, but whether voice agents can become responsive, interruptible, emotionally aware, and useful in real workflows.

On August 1, AGI House is co-hosting Voice Agent Build Day with Inworld, a full-day build session for founders, engineers, and researchers working on real-time voice AI.

Builders will work hands-on with Inworld’s real-time voice stack to create agents that can listen while speaking, handle interruptions, understand tone and emotion, and respond in under a second.

Internpalooza: Hot Takes, Lawn Games, and One Very Meta Agent Demo

Over 100 interns, founders, and student builders from MIT, Harvard, and a dozen other schools signed up to spend a Saturday afternoon with us. Pool, food trucks, lawn games, and a roving mic asking three questions:

• What’s your hottest take on AI right now?

• If you had unlimited funding, what would you build?

• What’s your personal definition of AGI?

We heard everything from bold predictions to definitions of AGI we had never considered.

The most meta moment: AGI, Inc. demoed a mobile agent booking a ride to AGI House end to end. We asked everyone what agents might change, and then one quietly handled the logistics.

If you were there, tag us and share your favorite moment, or the hot take you did not get to say on camera.

Huge thanks to TNT and AGI, Inc. for co-hosting and helping bring everyone together.

AGI House Talent Board

We’re starting a talent campaign to help high-signal builders discover roles across our portfolio and extended network.

Our AGI House Ventures portfolio startups are hiring:

TenX Semi — building AI-powered agents and tools for chip design.
Open roles: AI Systems Engineer, AI Application Engineer, Head of AI
Strong fit for engineers with experience in LLMs, AI agents, code generation, RAG, document understanding, RTL, verification, or chip design workflows.

Clair Health — building the world’s first continuous, non-invasive hormone monitor.
Open role: Senior Full-Stack Engineer.
Strong fit for engineers with TypeScript/Node.js, PostgreSQL, backend systems, Kubernetes, AWS, or privacy-conscious health data experience. View all roles and full descriptions here.

If you’re interested in any of the roles, or know someone who would be a strong fit, feel free to email [email protected].

Stay Updated with Us