Executive summary
SF Tech Week Oct 6–12 hosts over 1,100 events. 380+ focus on AI. That density is the signal. The center of gravity is clear: agentic systems, MCP interoperability, realtime voice, infra scale, plus a tougher stance on reliability, evals, security, and policy. Expect fewer science-fair demos, more production talk, and concrete vertical wins. Use this week to set Q4–Q1 direction: where agents fit in to operating models, how to measure value beyond pilots, which infra decisions to lock, and how policy will reshape risk postures.
Event landscape at a glance
Of 380+ AI-relevant sessions, the mix skews applied, with strong showings in infrastructure, security, and agents.
Event type distribution:
Networking & Social: ~120
Panels & Roundtables: ~60
Workshops & Sessions: ~50
Dinners: ~40
Pitch Competitions: ~20
Hackathons: ~15
AI Other (runs, yoga, sports, birds-of-a-feather): ~76
Thematic breakdown
Vertical AI —————— 25% ▉▉▉▉▉▉▉▉▉▉▉▉▉
Infra & Security ————- 21% ▉▉▉▉▉▉▉▉▉▉▉
Policy & Economics ———– 21% ▉▉▉▉▉▉▉▉▉▉▉
Agents, MCP, Orchestration — 17% ▉▉▉▉▉▉▉▉▉
Foundation Models & Research — 8% ▉▉▉▉
Multimodal & Voice ———– 8% ▉▉▉▉
Seven trends to track
1) Agents move from proof to process
Teams are showcasing handoffs, policies & evals that let agents drive measurable workflows. The shift is from “agent can” to “agent should,” with controls, observability & SLAs. Monday, Oct 6 alone schedules 48+ gatherings; what surfaces isn’t model worship but production theology with engineers comparing what breaks at scale & CTOs mapping gaps in the agent stack. The question has shifted from “can we build this?” to “how do we keep this running?” Budget implication: fund evals, logging & policy tooling alongside feature work.
2) MCP as connective tissue
Model Context Protocol shows up across hackathons, panels & demo nights. The value isn’t just tool access; it’s governance, routing & shared mental models for multi-tool agents. Four events spotlight MCP within 72 hours (OpenAI, Vercel, Langflow, Stytch; WorkOS demos; Apollo x GraphQL; Leanmcp hackathon). Interoperability is the battlefield: whoever sets context-passing standards shapes how agents communicate. Early infra choices compound; watch auth patterns, directory growth & enterprise connectors.
3) Real-time voice becomes table stakes
Voice moves from novelty to default for assistants coordinating people, data & systems. Expect focus on latency, barge-in, diarization & human-in-the-loop handoff.
4) Infrastructure realism replaces GPU euphoria
Cooling, power, placement, observability, evals & cost curves matter as much as raw FLOPs. Compute topology now dictates competitive position; building data centers becomes building moats. C-suite takeaway: infrastructure choices are strategy. Lock a view on placement, partners & cost discipline before 2026 commitments.
5) Security shifts left for agents
IR, retrieval security, identity & guardrails are non-negotiable. Expect more “do not break in prod” sessions with concrete incident patterns. Require control points for prompt injection, tool abuse, data exfiltration & replay.
6) Vertical depth beats generality
Housing, healthcare, fintech risk, energy & bio show credible paths to value with domain data, UX & regulatory context. Instead of horizontal tools adapted vertically, models train from domain ground truth. Advantage accrues to proprietary datasets in regulated industries. Vertical AI isn’t niche; it’s capture. Ask for line-of-business scorecards, not just platform roadmaps.
7) Policy, law & economics step onto the main stage
Legaltech, M&A policy & State of AI give leaders the context for procurement, compliance & capital planning. Translate findings into contract clauses, RFP language & board questions.
What the Calendar Reveals
This week marks a shift from potential to production, demos to deployment, research flex to revenue. Agents dominate because they’re AI’s operational form. MCP matters because interop determines ecosystem winners. Infrastructure commands attention because scale is now the product.
Where talent congregates reveals what’s valued. Repetition across venues signals emerging standards. Hosts indicate power concentration. OpenAI anchors with DevDay. Anthropic’s Ben Mann curates a founder salon. Between them: breakfasts on physical AI, panels on data-center economics & roundtables on agent handoffs.
What remains unclear: whether this acceleration yields sustainable systems or spectacular collapse. The Coming AI Winter (Wed, Pebblebed) suggests the question is live: infra costs, energy demands & sustainability constraints. The counternarrative is scheduled, not featured.
For executives calibrating strategy, the map precedes the territory, but this calendar may be the most honest map available.
My top 5 picks
1) OpenAI DevDay 2025
Mon Oct 6, 9:00 AM–6:00 PM, San Francisco devday.openai.com/2025
Leading AI developers and experts unveiling the latest innovations from OpenAI, with in-depth technical sessions and forward-looking presentations that will shape the near future of product and platform decisions.
2) Anthropic AI Founder Salon
Tue Oct 7, 2:00–6:00 PM partiful.com/e/hUd3ekBEfQG0ZIfWtqj4
Host: Benjamin Mann**, Co-founder and Technical Lead, Anthropic. Guests include:** Jeff Wang (CEO, Windsurf), Matt McClernan (CEO, Augment Code), and Eric Jing (CEO, Genspark). What to expect: deep-dive sessions on Claude capabilities and new features, lessons on building and scaling AI companies and products, the latest research updates from Anthropic, and direct access to Anthropic’s Startup and Research teams.
3) MCP and AI Agents: Panel with OpenAI, Langflow, Vercel, Stytch
Thu Oct 9, 5:00–8:00 PM, San Francisco * luma.com/hmq9j8n3
Expert panel on agent and MCP infrastructure, architecture, and emerging standards. We will open with a panel and dig into agent infrastructure patterns that are working and breaking in production, the latest developments in MCP and implementation best practices, new frameworks changing day-to-day agent building, and how standards are coalescing. Panelists: Dominik Kundel (Developer Experience, OpenAI), Andrew Qu (Office of the CTO, Vercel), Carter Rabasa (Developer Relations, Langflow). Moderator: Julianna Lamb (CTO and co-founder, Stytch). Who this is for: engineers, engineering leaders, product leaders, and anyone exploring MCP and agentic AI.
4) State of AI 2025 — Launch Event
Thu Oct 9, 5:00–8:30 PM, San Francisco * luma.com/soai
Nathan Benaich unveils the tl;dr of the 2025 State of AI Report with notable guests. Speakers include Sholto Douglas (Anthropic, xDeepMind) on scaling RL and on Economic Index with HCI, Joseph Spisak (Meta), and Ali Madani (Profluent) on frontier AI for science.
5) Adapt or Be Automated: Managing Change in an AI World
Thu Oct 9, 5:30–8:30 PM, San Francisco * partiful.com/e/50uM1bUpP8GBak8QRBbX
Panel on how enterprises, startups, and investors manage change during AI-driven transformations, balancing innovation, risk, scale, and impact. The conversation covers boardroom strategies, funding bets, ground-level execution, and ethical implementation with tensions, tradeoffs, and opportunities shaping adoption. Panelists: Natalie Han, PhD (VP and Chief Product Officer, GenAI at Business AI, SAP), Lauri J. Moore (Partner, Bessemer Venture Partners), Shirshanka Das**(Co-founder and CTO, DataHub),** Alistair Barr (Tech Anchor, Business Insider). Moderator: Domitilla Caputo (SF, LA, SD) (Strategic AI Partnerships, SAP).
