GitHub Universe 2026 says this year's pitch is that builders become orchestrators, and the in-person pass is already sold out. That slogan turns out to be less a marketing flourish than a citywide mood board. The unofficial fringe around Universe week in San Francisco keeps circling one question: once the agents are coding, who exactly owns the consequences?

On Monday night, exe is hosting Are We Running Too Many Agents? at Sentry, which is an unusually honest title for a week otherwise full of product confidence. The next day, incident.io brings its SEV0 reliability conference to SFJAZZ, while Rootly AI gathers Replit, NVIDIA, and WorkOS for a panel on Running Production with Agents. The promo copy does not bother pretending the hard part is generation. It is trust, investigation, code volume, and the small administrative matter of who notices when the robot has ideas.

Even the more aspirational rooms are singing from the same hymnal. OpenRouter and Cline want to debate who controls the future of AI coding. Factory, Sapienne, and Sequoia are running a Women in AI night where newcomers can bring a laptop and learn to delegate engineering work to agent Droids. The Syntax meetup rolling into town ahead of Universe is agenda-free, but even that functions like a weather report: everybody wants to compare notes before the keynote circus starts.

The scene signal is deliciously clear. San Francisco still loves a big tent event, a sold-out badge, and a beer garden full of builders. But the side-quest schedule says the city has moved past pure demo lust. The hot conversation is no longer can agents write code. It is who reviews the output, who runs prod, and whether orchestration is just a fancy word for managerial anxiety with better snacks.