Research
Where behavioral AI engineering is going.
The Question We Are Trying to Answer
Behavioral AI engineering, as it exists today, assigns agents behavioral profiles drawn from established frameworks: Myers-Briggs types, Clifton Strengths patterns. This produces meaningfully better outputs than systems with no behavioral modeling. Diverse agent profiles debate longer, catch more, and arrive at better solutions.
But a profile is still an approximation. It captures type, not individual. The more interesting question — the one that motivates our research — is whether AI agents can be built to model how a specific person actually makes decisions.
Not a behavioral type. Not a personality category. The actual decision patterns of a specific individual: what they weight, where their judgment is strongest, what they tend to overlook, how they reason under uncertainty.
Wouldn't you like an AI agent that makes decisions the same way you do?
What This Is and Is Not
This is frontier work. It is not a product. There is no timeline we are prepared to commit to publicly, and we are not going to suggest otherwise.
What we can say is that it is grounded. The behavioral engineering service we deliver today — using established behavioral frameworks to design agent ensembles that produce better outputs through genuine deliberation — is not a stepping stone built for marketing purposes. It is the foundation the research is built on. We understand how behavioral modeling affects agent output because we do this in production, for real clients, on real problems.
The distance between "agents grounded in behavioral types" and "agents that model individual decision-making" is substantial. We know what it would take. We are working on it because we believe it is achievable and because the work underway is different from what anyone else is doing.
The Foundation
The research builds on demonstrated results from the behavioral engineering work we do today.
Agents modeled on distinct behavioral profiles produce substantively better outputs than homogeneous ensembles. This is demonstrated, not theorized. The diverse teams analogy holds: disagreement is not a failure mode in a well-designed ensemble, it is the mechanism that produces better outcomes.
Individual decision-making patterns are more stable and more predictable than most people assume. The frameworks we use are not personality tests in the pop-psychology sense. They are models of how people process information and make judgments, with decades of research behind them.
The path from behavioral type to individual model is a research problem, not an engineering problem. That distinction matters. We are not waiting for a component to be built. We are doing the underlying work that would make the component meaningful.
Follow the Work
We share findings and thinking as the research develops. No newsletter. If you want to follow the work, LinkedIn is the place.