Every wave we've lived through - internet, cloud, mobile, social - was a revolution in communication. Faster pipes for distributing information. AI is a different shape of wave: a revolution in computation. It changes how information is processed, not how it's moved.
That sounds like a semantic distinction. It isn't. Communication revolutions speed up the existing economy. Computation revolutions rewrite which jobs can exist.
Sequoia's 2026 keynote made the case that we just crossed a hard break. Three inflection points: ChatGPT in November 2022, the o1 reasoning model two years later, and Claude Code / Sonnet 4.5 / Opus 4.7 over the last few months. The first two were continuous progress. The third one is discontinuous - long-horizon agents that can dispatch on a job, fail, recover, and persist until done. By any practical, commercial, functional definition, that's AGI. You don't have to call it AGI. You do have to notice that the ground moved.
For most of the last few years, AI applications were faster horses. They made you 10 to 40 percent more productive. Now we're getting cars. Applications that make you 10 to 40 times more productive - and change the shape of the work entirely. Nathan from Zed shipped a three-year moonshot over Christmas with Claude Code. Brett Taylor rebuilt Sierra over a weekend. Notion rewrote 8 million lines of code in six weeks.
Whatever you imagined building over the next hundred years is now possible in a hundred days. That's not hyperbole; that's the operating reality at the labs.
Services is the new software
Here's the part that should reorder how every founder thinks about TAM. The first 15 years of the cloud transition grew the global software market from $350B to $650B. Services - legal, medical, accounting, financial, marketing - sits at roughly $10 trillion. US legal services alone is $400B, the same size as the entire software industry, in one vertical and one geography.
Until now, services was off-limits to software. You couldn't sell software to do litigation; you had to hire lawyers. With agents, that changes. You can hire an agent to negotiate a contract, file your taxes, run a generative media campaign, or inspect your genome and prescribe medication. Humans cost salaries. Agents cost tokens. Tokens are cheaper than salaries today, and the gap is widening.
This isn't "SaaS is dead." It's the opposite. The tools SaaS built over the last two decades - Slack, terminal, browser, file system - turn out to be perfectly portable to agents. The value of those tools explodes as agents start using them at scale. But the new TAM, the trillions, sits in the services layer that agents can finally touch.
The MAD playbook for building on top of the labs
If you're building on foundation models, the framework for not getting flattened is MAD: Modes, Affordance, Diffusion.
Modes. The capabilities under you change every week. Your customers don't. Wrap yourself around the customer, not the model. Best product still wins, but in a world where product capability turns over monthly, customer intimacy is the durable wrapper.
Affordance. A hammer doesn't need a manual. Claude Code is staggeringly powerful - and the average Fortune 500 employee can't get past the terminal. Your job is to take frontier capability and create paths of least resistance for a specific customer with a specific problem. Make the powerful thing braindead simple to use.
Diffusion. The rate at which capabilities ship from the labs is wildly faster than the rate at which they diffuse into the average enterprise. That gap is your opportunity. Every day the labs move faster than the market adopts, the gap gets bigger. Live in that gap.
And before anyone gets too settled - no lead is safe. You can't pass fifteen cars in the sun. You can in the rain. There's a torrential downpour of new capability coming out of the labs every week. Anybody can win. Nobody is safe.
Aluminum, alien design, and the part nobody is ready for
Aluminum was once the most precious metal on earth. The cap of the Washington Monument was 100 ounces of it, on display at Tiffany's because it was rarer than gold. Then a young inventor cracked electrolysis. Within a few decades, aluminum was wrapped around sandwiches and tossed in the trash.
Aluminum is intelligence. Electrolysis is AI. PhD-level skills that took decades to acquire are about to be invoked instantly and discarded just as fast. That should terrify anyone whose moat is credentialed expertise. It should excite anyone whose moat is anything else.
A second consequence: the world starts to look weird. NASA handed antenna design to an evolutionary algorithm in 2006. The result was wildly more efficient - and looked like nothing a human would draw. When AI does the cognition, the output doesn't have to make sense to us. AI-designed chips, buildings, drugs, businesses are going to look alien. Get comfortable.
What's still ours
Cognition is on its way to being mostly machine work. 99.9% of it, eventually. That's the trajectory. The honest answer to "what's left for us" isn't a productivity tip; it's relationship. The reason anyone shows up to a keynote, signs a deal, picks one founder over another, or stays loyal to a brand isn't going to be the cognitive labor - that's commodified. It's going to be the human connection.
That's not a soft observation. It's the part of the moat that AI can't take.
Source: The Cognitive Revolution: Sequoia AI Ascent 2026 Keynote
No comments:
Post a Comment