Research
Notes on the behavioral layer.
Technical write-ups and design notes from Humalike on how the four behavioral primitives — turn-taking, social norms, emotions and social cues, relational memory — get composed into a runtime that lets AI agents behave like humans in real human scenarios.
Research posts ship here as we publish. The first ones are queued — listed below.
Queued
Coming soon
Turn-taking in multi-party AI agents
How HUMA decides when to speak, when to wait, and when to interrupt — and why this is a different problem from "next token prediction."
Coming soon
Relational memory vs. context windows
What an agent should remember about a person across sessions, what it shouldn't, and why bigger context windows don't close the gap.
Want a heads up when these publish? Reach out through the contact form on the homepage.