2026: This is AGI
Sequoia by Pat Grady and Sonya Huang
Sequoia says AGI is already here, and the first real proof is coding agents. The big bet: by 2026, agents won’t just chat — they’ll work for hours like people do.
Based on reporting by Sequoia, by Pat Grady and Sonya Huang — 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 is making a blunt claim: AGI is no longer a future milestone, but a present-tense reality. The firm isn’t trying to settle the philosophy seminar version of the term. It’s using a practical test instead — can a system figure things out on its own and get useful work done?
That’s why the essay lands on coding agents first. Not because they’re cute demos, but because they already show the three ingredients Sequoia cares about: pre-training, reasoning at inference time, and long-horizon iteration. ChatGPT brought the knowledge. o1 brought longer thinking. Claude Code and other coding agents, in Sequoia’s telling, added the ability to keep going, recover from mistakes, and pursue a goal without constant nudging.
The example it gives is very specific. A founder asks an agent to find a developer relations lead for a platform-focused company. The agent doesn’t stop at LinkedIn titles. It searches conference talks, checks Twitter activity, filters for real engagement, notices who’s gone quiet lately, and narrows the field to one person whose background, timing, and online behavior all line up. Then it drafts a tailored outreach note. Total time: 31 minutes.
Sequoia says this is what it means to “figure things out” — not following a script, but moving through dead ends and changing course. It also admits the systems still fail. They hallucinate. They lose context. They can charge off in the wrong direction. But the firm argues the failures are getting fixable, and the trend line is what matters.
That trend line matters because the next step isn’t just better chatbots. Sequoia thinks long-horizon agents are scaling fast enough that tasks taking human experts a day could be routine by 2028, a year by 2034, and a century by 2037. Its conclusion for founders is simple: the AI products of 2026 and 2027 won’t feel like tools. They’ll feel like colleagues.
My take — AI-written commentary, not fact-checked reporting
Sequoia is right about one thing the hype crowd keeps dancing around: chat was never the endpoint, just the demo reel. The real shift is when software stops answering and starts finishing. That’s the part that should worry every founder still selling “assistants” like it’s 2023.
Read more about this at: Sequoia
