Intelligence Architect

Basis builds real agents that do real work in the real economy. Our agents operate for hours at a time, performing end-to-end work for some of the largest accounting firms in the world.

We recently raised $100M at >$1B valuation and are racing to deploy the most advanced applied ML at production scale.

Our investors include: Khosla Ventures (Keith Rabois & Vinod Khosla), Accel (Miles Clements), Google Ventures, Nat Friedman & Daniel Gross, Adam D'Angelo, Jeff Dean, Jack Altman, Noam Brown, Kyle Vogt, Amjad Masad, Clem Delangue and many other operators/technical leaders.

"Basis is on the frontier of building production-grade, long-horizon agents. They've pushed the limits of what we thought our models could do on real-world, economically valuable, complex accounting tasks. They've been a great collaborator in helping us shape what the future of agents looks like." — Prashant Mital, Applied AI Lead, OpenAI

Every team building agents has access to the same models. Most of the attention goes to the product and its behaviors. The compounding work sits underneath: what the agent knows, how it reasons about its work, and how we measure whether it did the job well.

At Basis, our agents autonomously perform real accounting work: reconciliations, tax preparation, audit procedures, financial analysis. Often, the quality of this work is not determined by which model we use. It is determined by the context those agents receive: the procedures that define what to do, the domain knowledge that defines how, and the evaluations that tell us whether it worked.

To deliver frontier intelligence, we treat context like a product. The procedures, the skills, the system prompts. Each is owned, versioned, and held to a quality bar.

Intelligence Architects build the coherent systems that construct our agents’ intelligence from disparate information. You hold the mental model of what each agent knows, how it reasons, and how we measure whether it's working. You write the procedures and skills. You define what "good" means at every layer and design how we measure it. Ultimately, you're orchestrating an intelligence whose ceiling rises with every model release, and whose capability compounds with every edit you make.