Agentic Systems Lab

An agentic system is one that pursues goals over extended horizons: decomposing objectives into plans, using tools, maintaining memory, and recovering from failure , without a human approving every step. Where a model responds, an agent acts.

The lab studies what happens when single-model intelligence becomes a working society of agents: how tasks are decomposed, how context is preserved, and how multiple agents coordinate without collapsing into noise.

Most real work is long-horizon. Single-shot intelligence , however capable , cannot run a supply chain, conduct a literature review, or operate a business process. The gap between what models can do in one turn and what organizations need done over weeks is the defining engineering problem of this decade.

Coordination is the harder half of the problem. A society of agents that cannot delegate, verify each other's work, or resolve conflicting goals is worse than one good agent.

Develop the architectures, protocols, and evaluation methods that make autonomous agent societies reliable enough to trust with real work.