Issue #25Rally SF Newsletter

Monday, October 5, 2026

It's SF Tech Week, which this year means about 1,500 events between today and Sunday, split across Luma and Partiful. You can find them at rallysf.com/events, sorted by day and time. The picks below are the ones I'd choose if you build agents or robots.

I'll be at Encord's Long Horizon summit on October 20, the robotics day at Fort Mason. It opens with Deepak Pathak of Skild AI, Alex Kendall of Wayve and Quan Vuong of Physical Intelligence, three companies building general-purpose models for robots and cars. Code RALLYSF25 brings a day-one ticket from $399 to $299, and if your company pays for conferences, the site has a ready-made email for getting it approved. Rally earns a share of tickets sold with the code.

See you there.

Lee

Coming up at Rally

Rally Event

Builders Night

2026-10-12 · 6:30 PM - 9:00 PM

A weekly coworking session where builders show up with a project and leave with real momentum. Quiet deep work sprints paired with a share-back where you learn from others and get unstuck together.

Around SF this week

Agents in Production: Shared Memory (SurrealDB x Mastra) →

Monday, Oct 5, 5:00 PM - 8:30 PM

An agent that is right in the demo can answer confidently wrong once a fact changes underneath it. SurrealDB's Tobie Morgan Hitchcock and Mastra's Aamish Ahmad Beg cover where agent knowledge lives and who keeps it true.

Physical AI Operators #02 — The Data Layer of Physical AI @Robotics Center →

Tuesday, Oct 6, 6:00 PM - 9:00 PM

Robotics data moves from collection and training to deployment and commercial value, and this panel follows that path. Speakers come from AgiBot, Astera Robotics and Veesion, with networking before and after the panel.

Haunted Agent + Infra Horror Night @SF Tech Week →

Tuesday, Oct 6, 6:30 PM - 8:30 PM

Three infrastructure leaders get twelve minutes each to tell one true story of an AI or agent system going wrong. Matt Lenhard of Vectoral covers inference theft and token fraud, and NVIDIA's resilience engineer speaks from incident command.

The Model Decision: Optimizing Cost, Control, & Speed →

Wednesday, Oct 7, 5:00 PM - 8:30 PM

Ollama and Trajectory made different bets on closed models, open models and proprietary training, and each explains what they run today. Crusoe's Erwan Menard joins them on the cost, control and speed tradeoffs behind each choice.

Bay Area Frontier Research Club #26 | Agents in Production: The Data Layer — SF Tech Week (dinner + paper discussion @ Pebblebed) →

Wednesday, Oct 7, 5:30 PM - 8:30 PM

Agent memory has to stay valid as facts change and permissions shift, which makes it a data systems question. Zep's Daniel Chalef and UC Berkeley researchers Shu Liu and Shubham Agarwal present, with papers shared in advance.

Agents in Production: The Auth Stack for AI — SF Tech Week 2026 →

Thursday, Oct 8, 5:00 PM - 7:30 PM

Once an agent can reach Salesforce, Slack and internal APIs, someone has to decide its identity and what it may do. Agentic Fabriq and Open Future Forum host a technical discussion on least-privilege access, credentials and auditing.

Hard Negatives Meetup | Engineers Debate Night with Qdrant & Neo4j →

Friday, Oct 9, 6:30 PM - 9:00 PM

Engineers debate spicy topics like banning open weights in one-on-one matchups, and the audience votes live on each winner. This night with Qdrant and Neo4j is built for engineers, AI researchers and builders, with prizes for winners.

AI-Native Startup Hackathon →

Saturday, Oct 10, 10:00 AM - 8:00 PM

Teams build a startup in one day, from validating a problem to a working MVP and a judged pitch. AI teammates are built with Kylon, and judges score AI-native execution along with their creativity and effectiveness.

Robotics & World Models Reading Club 33: Perceiving, Predicting, and Planning for Physical Interaction + AdaJEPA. SF 10/10 →

Saturday, Oct 10, 2:00 PM - 5:00 PM

JEPA world models plan by predicting action consequences in a learned representation space, and NYU's Ying Wang presents AdaJEPA. Hongyu Li, a robotics researcher at NVIDIA, speaks on perceiving, predicting and planning for physical interaction.