CUGA is an open, configurable generalist agent harness built for real enterprise work — and it's ranked top-tier on AppWorld and WebArena, two of the most demanding agent benchmarks in the field. Join us for a live walkthrough with the IBM Research team behind it, plus demos across Health Insurance, Digital Sales, and Knowledge Management.
No registration required. Just save the link or drop your email to get a reminder.
Optional — drop your email and we'll send a calendar invite, a heads-up before we go live, and the recording afterward. No registration needed to attend.
Free · Live · Recording goes to everyone who opts in.
CUGA is built in the open, for the people building agents in production. In this Office Hours, the team behind it walks through what makes the harness ready for real work — and demos it across several enterprise domains.
Match the reasoning strategy to the task — not the other way around.
OpenAPI, MCP, and LangChain tools plug in cleanly through a single interface.
Compose and reuse capabilities across agents and teams.
Grounded, retrievable context — so agents act on what's actually true.
Coordinate specialist agents on complex, multi-step work.
Guardrails and approval flows that match how real organisations operate.
Agents that get better as they go.
Ship fast across multiple LLM providers.
You're evaluating agent harnesses for real workloads — not toy benchmarks. See how CUGA holds up.
You need policy, approval, and orchestration that fit how your organisation already works. See the layers that make CUGA enterprise-ready.
CUGA is open. Bring your own tools, skills, and integrations — and meet the team you'd be building alongside.
Nir leads the AI Agents group at IBM Research, focused on agent systems that work reliably outside the lab.
Free, live, and open — bring your hardest question.