Sequoia AI Ascent 2026: This is AGI

What the video covers

This keynote from the Sequoia AI Ascent 2026 conference features partners Pat Grady, Sonya Huang, and Konstantine Buhler making a case for why 2026 represents a practical threshold in AI development — not a philosophical one. The argument is that AGI is already here in the sense that matters: AI systems can now “figure things out” autonomously, sustaining multi-step work across extended periods, correcting their own errors, and completing goals without continuous human direction.

The keynote focuses on what Sequoia calls long-horizon agents — AI systems capable of working on a task for hours or days, backtracking when they hit dead ends, and delivering a finished output rather than a partial response. This is distinct from the AI behavior most product teams shipped in 2023 and 2024, where the model responded to a single prompt and required the human to stitch outputs together across sessions.

Who it’s for

Product managers and product leaders making roadmap decisions about where to invest in AI. The keynote is not a technical tutorial — it frames a strategic argument about the maturity of the technology and what that maturity means for product teams. Practitioners who are already deep in AI product development will find the framing useful as a way to calibrate their thinking against how major investors are reading the moment.

Key takeaways

1. The category shift from talkers to doers. Grady and Huang draw a clear line between what AI did before 2026 and what it does now: “The AI applications of 2023 and 2024 were talkers, but the AI applications of 2026 and 2027 will be doers.” This is a useful frame for product teams evaluating whether their current AI features are still in the conversation-assistance mode or have moved into task execution.

2. Users become managers, not operators. The keynote describes a shift in how people will interact with AI-powered products: rather than guiding a model step by step, users will assign goals and supervise outcomes — “going from working as an IC to managing a team of agents.” Product teams building AI features need to decide which mode their product is in and whether they are designing for the older or the newer interaction pattern.

3. Long-horizon reliability is the threshold change. The key capability change is not the model’s intelligence in isolation but its ability to sustain coherent, reliable work across many steps without failing or drifting. Sequoia treats this as the moment the technology crosses from impressive demo to deployable infrastructure.

4. The bottleneck is human judgment, not AI output. As AI takes on more execution, what becomes scarce is the quality of human direction — goal-setting, evaluation criteria, error detection, and strategic framing. Andrej Karpathy’s talk at the same conference reinforces this from the engineering side: “you can outsource your thinking, but you can’t outsource your understanding.”

5. Enterprise adoption is accelerating but governance is lagging. Data cited at the conference shows nearly 30 percent of Fortune 500 companies are paying customers of AI startups, but only one in five has a mature governance model for autonomous agents. This gap matters for product teams: the demand is real, but the organizational infrastructure to support agentic products at scale is still being built.

Worth watching if…

You are a product manager or leader making decisions about whether and how to build agentic AI into your product, and you want to stress-test your assumptions against how experienced investors are framing the current state of the technology.