QCon AI Boston 2026’s schedule is live, and the talks are clustered around the stuff people actually care about right now: agents in production, inference costs,.
Seeing “inference costs” in the headline feels honest — that’s the first wall you hit once the demo glow fades and someone asks for a budget.
I’m curious whether the “agent reliability” talks get specific about user-facing failure modes like silent wrong actions or messy handoffs, not just benchmark scores.
