Agentic Loops Happy Hour: Making Agents Reliable in Production
The AI Conference wraps Day 2, and the real conversations are just getting started.
"Loop." "Harness." "Eval." Everyone at the summit used these words, each meaning something a little different. This is the room where we pin them down.
The demo works once; the product has to work 10,000 times. So when an agent delivers, the real question isn't how capable the model is. It's this: who's actually in the loop, the model, the harness, or the human quietly catching the failures?
Leading it: moderated by Roan - GMI's Dev-Rel. Come compare notes and leave with a sharper picture than the buzzwords you walked in with.
What we'll get into
Where reliability really comes from: the model, the harness, or the human
What a benchmark score does and doesn't predict in production
Where self-correction earns its cost, and where it just adds latency
Schedule
🍻 6:00 pm · Doors: drinks, bites, games, and the agent-fails wall
💬 7:30 pm · Panel + live Q&A
🍸 8:15 pm · Open networking, raffle for GMI credits at 9
Who should come
Founders, engineers, and researchers building agents that have to survive real users. Around 80, in a room small enough to actually meet them.
About GMI AgentBox
Infrastructure for teams building reliable agentic workflows, from GMI Cloud.
RSVP to secure your spot; capacity is limited.