Prof. Liam Carter

Liam Carter grew up in Glasgow and learned to program by writing mods for video games, which he still considers the ideal introduction to software engineering. 'You have an immediate feedback loop,' he says. 'If the behavior is wrong, you know right away, and you fix it.' This philosophy of fast iteration and concrete verification runs through everything he builds.

His PhD on distributed autonomous systems gave him a rigorous foundation in the problems that arise when many independent agents must coordinate without central control — problems that turned out to be directly relevant to modern multi-agent AI systems. He arrived at DeepMind just as the lab was scaling its agent research and spent five years working on the systems that made large-scale experiments possible.

At Waymo, he led the team responsible for the engineering infrastructure that supports autonomous vehicle decision-making. He is fond of saying that 'an autonomous vehicle is just a very consequential AI agent,' and that working on safety-critical autonomous systems gave him a rigorous intuition about what it means for an AI system to behave reliably.

At Berkeley, he runs a research group focused on the systems engineering of reliable AI. He teaches a course on agent architectures that is consistently oversubscribed and has a waiting list of students from across engineering and computer science.