Cost per correct answer
Dinner. Unit economics of agents once the demo ends: routing, caching, distillation, and when to own the GPUs.
San Francisco
Winter 2026
A community for agentic builders
Small rooms for people who run inference for a living.
Off-the-record dinners and salons for CTOs, VPs of Engineering and Chief AI Officers putting agents into production. No panels. No pitches. Just tokens, latency and truth.
$ infer --model community-v0 --temp 0.20 tok/s
▍
12
seats per room
0
slide decks
1
rule: Chatham House
∞
context, not windowed
Upcoming forward passes
The p99 is the product
Salon. Latency budgets for multi-step agents: speculative decoding, batching, prefill vs decode, TTFT.
New York
Early 2027
Evals nobody publishes
Closed-door. How AI leaders actually decide a model is ready to ship. Chatham House rules.
Virtual + London
2027
Who's in the room
- → CTOs whose inference bill has its own board slide
- → VPs of Engineering on-call for agents at 3am
- → Chief AI Officers translating evals into strategy
Request a seat
Invites go out in small batches. Low temperature, high signal.