9:00–9:45 AM PDT · Main Stage
Opening keynote: the year agents grew up
Where the last twelve months actually landed, and what shipped.
BUILD YOUR DAYStar your sessions,
Star your sessions,
export your schedule.
No account needed — star sessions to build My schedule on this device, then export it to your calendar. All times PDT.
Friday Oct 16
10:00–10:30 AM PDT · Main Stage
Caching at the edge without losing your mind
What a CDN can and cannot do for an inference path.
10:30–11:00 AM PDT · Main Stage
Evals that survive contact with users
Offline scores lie. Here is what we replaced them with.
11:00–11:45 AM PDT · Workshop A
⚠ Demo conflict A — Platform deep dive
Double-booked into Workshop A on purpose, so the Conflicts tab has something to resolve.
11:30 AM–12:15 PM PDT · Workshop A
⚠ Demo conflict A — Vector search at scale
The other half of the room clash.
1:00–1:30 PM PDT · Workshop B
Retrieval is not a database problem
Before: after.
2:00–2:45 PM PDT · Studio
⚠ Demo conflict B — Agent evals live
Same speaker as the Guardrails workshop, at the same time, on purpose.
2:15–3:00 PM PDT · Atrium
⚠ Demo conflict B — Guardrails that do not annoy anyone
The other half of the speaker clash.
Saturday Oct 17
9:30–10:00 AM PDT · Main Stage
Cost controls for long-running agents
Budgets, caps and the alert that actually pages someone.
10:15–10:45 AM PDT · Workshop A
Observability for prompt pipelines
Traces that survive a prompt rewrite.
11:00–11:30 AM PDT · Workshop B
Scaling human review
What we learned staffing a review queue that never emptied.
4:00–4:45 PM PDT · Main Stage
Closing panel: what we would build next
Two speakers, one room, no conflicts.