AI Ascent 2026
Sequoia by Team Sequoia ● Covered by 2 sources
Sequoia says AI has moved from theory to real products, with agents as 2026’s big theme. Its big claim: the gap between model power and what companies use is still wide.
Based on reporting by Sequoia, by Team Sequoia — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Sequoia’s fourth annual AI Ascent took over San Francisco on April 20, drawing more than 150 founders and researchers from across the field. The guest list included Demis Hassabis, Andrej Karpathy, Greg Brockman, Boris Cherny from Anthropic, Dmitri Dolgov from Waymo, and Jim Fan from Nvidia. This was not a quiet meetup. It was a statement of where the money, talent, and ambition are gathering.
Pat Grady opened with the bluntest line of the day: AI is not a tweak on the old world, it is a new kind of computing. His advice to founders building on top of model labs was equally blunt. Build moats from the customer backward. Design for affordance. And lean into the gap between what the models can already do and what most Fortune 500 companies have actually deployed.
Sonya Huang pushed the strongest theme of the event: 2026 as the year of agents. Her case rested on three ingredients that have finally come together — models, tools, and harnesses. That’s the sort of sentence that sounds simple until you realize how much of the industry has spent the last few years missing one of those pieces.
Konstantine Buhler widened the frame even further, arguing that the cognitive revolution will rhyme with the Industrial Revolution, only faster and bigger. The program then ran through a stack of big, slightly wild topics: the long arc of agents, robotics, data centers in space, data efficiency, and new science around neural networks. That mix says a lot. The field is no longer asking whether AI matters. It’s arguing over how far the machinery can spread, and how fast humans will reorganize around it.
Sequoia posted a selection of videos from the event, with the full lineup available in its YouTube playlist.
My take — AI-written commentary, not fact-checked reporting
This is the right kind of AI event: fewer victory laps, more people admitting the gap between what models can do and what companies actually ship. The real story isn’t bigger demos, it’s the boring plumbing of tools, harnesses, and customer fit — the part where hype goes to get a job. The industry keeps selling magic; the winners will be the ones who sell something that survives contact with procurement.
Read more about this at: Sequoia
