Berkeley Agentic AI Summit 2026: Recap, Speakers, Recordings
August 1-2, 2026
The Berkeley Agentic AI Summit 2026 wrapped August 1-2 on the UC Berkeley campus, hosted by Berkeley RDI (Center for Responsible, Decentralized Intelligence). The 2026 edition drew 5,000-plus in-person attendees and...
The Berkeley Agentic AI Summit 2026 wrapped August 1-2 on the UC Berkeley campus, hosted by Berkeley RDI (Center for Responsible, Decentralized Intelligence). The 2026 edition drew 5,000-plus in-person attendees and 200-plus speakers across four stages, including Andrej Karpathy, Wojciech Zaremba, Andrew Ng, and Ali Ghodsi (checked 2026-08-24, rdi.berkeley.edu). Standard tickets sat at $499. Every session was livestreamed free, and Berkeley RDI historically posts full session videos to YouTube after the event.
I track AI events because the ones that grow this fast in one year signal a real shift, not a marketing bump. Berkeley grew from roughly 2,000 in-person attendees in 2025 to 5,000-plus in 2026 on the same single-topic focus, which matters more than one more general AI summit adding “agentic” to its keynote title.
The reason to care about this specific event is who was in the room. Berkeley systems and security faculty (Ion Stoica, Dawn Song, Sergey Levine) sharing a plenary with OpenAI’s co-founder, DeepLearning.AI’s founder, Databricks’ CEO, and general partners at a16z and Sequoia is a genuine cross-section of agentic AI right now. That is the mix the recap below is built around.
What the 2026 program actually covered
Berkeley RDI ran a plenary stage plus three named tracks: Atlas (foundational research and model capabilities), Nexus (developer platforms and agent frameworks), and Compass (enterprise applications and safety). Four concurrent stages meant no attendee saw everything, which was the point. The Compass track in particular was built around the “evaluating whether an agent is safe, reliable, and worth deploying” gap that Dawn Song’s presence flagged as serious, not token, content.
Session topics spanned agentic AI infrastructure, agent frameworks, robotics and world models, enterprise applications, agent evaluation and benchmarks, AI security, plus finance, legal, healthcare, and scientific discovery use cases. A Startup Spotlight pitch competition, hands-on workshops, and 200-plus poster presentations rounded out the two-day main program, with side events running July 30 through August 4.
Speakers who anchored the 2026 plenary
The plenary lineup mixed Berkeley research faculty with frontier lab builders and venture capital. Here is the confirmed short list Berkeley RDI and the Haas School of Business coverage named ahead of the event.
| Speaker | Known for |
|---|---|
| Andrej Karpathy | AI researcher, founder of Eureka Labs |
| Wojciech Zaremba | Co-founder, OpenAI |
| Andrew Ng | Founder, DeepLearning.AI |
| Ali Ghodsi | Co-founder and CEO, Databricks |
| Sergey Levine | Robotics and reinforcement learning, UC Berkeley |
| Ion Stoica | Distributed systems, UC Berkeley; co-founder, Databricks |
| Dawn Song | AI security research, UC Berkeley |
| Pushmeet Kohli | VP of Science, Google DeepMind |
| Martin Casado | General Partner, a16z |
| Alfred Lin | General Partner, Sequoia Capital |
The full lineup ran 200-plus speakers, spanning academia, frontier labs, enterprise AI teams, and venture capital. Casado and Lin sitting on the same agenda as Berkeley systems researchers is the clearest signal of what this event actually is: the point where agentic AI research meets the capital deciding which of it gets funded next.
Pricing in 2026 as a 2027 baseline
Berkeley RDI ran a simple two-tier structure with a hard early-bird cutoff. Student tickets sold out well before the standard tier, which is worth planning around for 2027.
| Ticket | 2026 price | Notes |
|---|---|---|
| Early Bird | $399 | Closed July 5, 2026 |
| Standard | $499 | Ran July 6 through the event |
| Student | Sold out | Gone before standard opened |
| Virtual Livestream | Free | Global, all sessions |
You should treat these numbers as the 2027 planning baseline, not a forecast. Berkeley RDI has not announced 2027 dates or pricing as of 2026-08-24. The student tier selling out first is the pattern worth remembering.
Where the recordings live
If you missed the event, the free livestream is what to check first. Every session ran live during the summit through Berkeley RDI’s Luma page, and Berkeley RDI posted full session videos from the 2025 Agentic AI Summit to its YouTube channel after that event. On-demand replays for 2026 are a reasonable expectation on the same pattern, but Berkeley RDI has not published a formal on-demand policy I could verify for 2026 yet.
If you want the plenary keynotes specifically, subscribe to Berkeley RDI’s YouTube channel and set a notification. That is the surest signal for when the replays land, more reliable than checking the summit’s Luma page after the fact.
Who this event was for, and who should skip 2027
I get asked this every year when a large AI event wraps, so here is the honest sort. Berkeley Agentic AI is a working-practitioner and research event, not an AI 101 course, and the program assumes you already know what an agent framework, a tool call, or an evaluation harness is. That was clear in the 2026 program and will stay clear in 2027.
The right fits for 2027 are AI engineers building agent pipelines, ML researchers working on evaluation and safety, venture investors sourcing deals in the space, and policy people who want to understand autonomous systems technically before regulating them. If you are new to AI generally and looking for a beginner-friendly overview, this is the wrong room and the free livestream will not fix that.
For teams weighing whether to send four people to 2027, my rule is send one senior engineer in person for the hallway conversations and workshop access, and have the rest of the team watch the livestream for the plenary and main-stage talks. Two days of $499 tickets plus Bay Area travel adds up fast, and the split covers the networking value and the content without doubling the budget.
What to do before Berkeley Agentic AI Summit 2027
Berkeley RDI has not announced dates yet, but the growth curve (roughly 2,000 in-person in 2025, 5,000-plus in 2026) means capacity will be a real constraint next year. If you plan to attend, watch the official Berkeley RDI event page for the 2027 announcement and register the moment early-bird opens. Student tickets in particular go first.
While you plan around it, I keep tracking the AI events that shape which tools show up in real engineering teams, and adjacent events like AgntCon + MCPCon and the AI Ethics and Society conference cover the same practitioner audience from different angles.