100 Years of Progress in 100 Days: Sequoia AI Ascent 2026 Keynote
In a Nutshell
Sequoia AI Ascent 2026 keynote declares AI as the biggest, fastest tech wave yet—shifting from communication revolutions (internet/cloud/mobile) to computation (agents achieving functional AGI via ChatGPT, o1 reasoning, and long-horizon models like Claude Code/Nopus), enabling 10-40x productivity and $10T services markets replacing software. Builders should adopt MAD strategy (customer-back Modes, intuitive Affordance, rapid Diffusion) amid exploding agent capabilities for coding, medicine, law, and more, compressing 100 years of progress into 100 days. Physical work is 99% mechanized; cognitive work follows suit via neural nets, birthing alien designs, emerging AI sciences like thermodynamics, and human art/forms of unreason that prioritize relationships over commoditized labor.
These notes were generated by AI and may contain inaccuracies.
Thank you all for being here. We do this as a service to the community because we are living through important times and it's an honor to serve as a gathering place. This is the best agenda and best set of attendees. Thank you for being here today despite being busy.
We have an exceptional agenda. Sonia, Constantine, and I will say a few words to start. We have the privilege of being in many conversations with interesting people, so once a year we synthesize and share what we've been hearing.
Zoom out to silicon-based transistors that gave this area its name, built into systems connected by networks, public as the internet, supporting applications like social media in the cloud, on mobile devices in our pockets, today capable of AI that is indistinguishable from magic.
All waves are additive. We needed decades of evolution for compute, bandwidth, data, talent to make the most of this moment.
This AI wave is different in three ways:
-
Biggest wave yet, specifically the first that is both software and services. Top row: first 15 years of cloud transition, TAM for software from $350B to $650B, cloud grew to $400B of that. Bottom row: new services revenue, ~$10T (could be $5T or $50T). Legal services in US alone is $400B market—one vertical, one geo, same as all of software. Opportunity is immense.
-
Fastest wave yet. White space on AI side filling fast. Logos are companies that hit $1B+ revenue from cloud, mobile, AI shifts. More coming soon at current pace.
-
Two kinds of tech revolutions: communication (information distribution: internet, cloud, mobile) vs. computation (information processing: AI). Floor keeps moving—technology foundation changes daily with new capabilities.
Three major inflection points:
- Nov 2022: ChatGPT—world saw power of pre-training.
- ~2 years later: o1 model—reasoning, second scaling law around inference-time compute.
- Recently: Claude Code, Nopus 45/47—power of long-horizon agents.
Discontinuous change between 2 and 3—this is AGI from commercial/practical/functional standpoint: dispatch agent to do a job, recovers from failure, persists until done.
Even if not AGI, cars have arrived—apps that make you 10-40x productive, change how you work, nature of work/organization. Previously: faster horses (10-40% productive, no fundamental change).
Don Valentine: "So what?"
Race has begun—different kind: driving/building a car ≠ riding/caring for horse. Nobody has all answers; gather to learn fast. $10T at stake commercially. Labs from tech-out, startups/customer-back (most here building on top).
Advice for building on labs: MAD (Modes, Affordance, Diffusion)—not actually mad/angry.
Modes (Customer-Back)
Merchandising cycle: links from idea to happy customer. Tech-out vs. customer-back approaches differ per link.
In computation revolution, look up at customers (changing slower than capabilities). Things you build might be irrelevant tomorrow; wrapping around customers more durable. Product/tech insanely important (best product wins), but go as customer-back as possible.
Affordance
From design: object with affordance doesn't need explanation (e.g., hammer—2yo knows to hit stuff). Claude Code powerful but low affordance (Fortune 500 employee struggles in terminal). Opportunity: build paths of least resistance for specific customers/problems—braindead simple to outcome.
Diffusion
Diffusion gap: capabilities created faster than diffusing to market (e.g., foundation models > Fortune 500 enterprise). Gap/opportunity grows daily for app layer.
No lead safe (racing: can't pass 15 cars in sun, but can in rain)—torrential downpour of capabilities means anybody can win.
Agents are what's happening now for 2026.
Flashback 2022: AutoGPT, Baby AGI—overnight GitHub hits. Took GPT-3, added tools/loop toward goal. Promising but failed repeatedly—useless. Models weren't ready.
Today: agents everywhere, actually working. Home runs: Claude Code (technical), OpenClaw/lobster brethren (democratizes to anyone with phone). Anybody can create agents.
Building agents for everything: silly (OpenCall snitches neighbors for tax fraud), entrepreneurial (generative media campaigns for construction), professional (Sequoia internal race for best agents to do jobs better).
Agent definition: system that perceives environment, chooses actions, progresses autonomously toward goal. (Made video in C-dance—video models advanced.)
Three functional components:
- Reason/plan: baseline intuition, think on fly.
- Actions: tools, search, write, compile.
- Iterate toward goal: persistence for long horizons.
Agency = ability to get done.
Components progressed:
- Models (brain): meter = sustain progress on complex task w/o derailing. From tens of minutes (year ago) to hours today—key advance for long-horizon tasks.
- Tools (arms/legs): terminal, iMessage, Slack, web search, computer use. 20yrs human tools port to agents. SaaS not dead—value explodes as agents use them.
- Harnesses (persistence): stay on task/adapt. Feedback loop cranks w/ RL (driving school, RL gyms: mechanical eng, design, finance). Early self-improvement (Andrej's project: GPT-2 level model in 2hrs).
Sliding scale:
- 2023: tab autocomplete (AI assists human)—incremental.
- Now: agentic development (human instructs/manages agent team)—transformative.
- Next: background/async agents (spawn sub-agents)—likely overtakes in volume (leverage).
- Frontier: dark factories (no human review)—seen in production (e.g., cybersecurity) w/ guardrails/eng.
Progress: helpers → managed interns → self-managing → prod w/o oversight. Across all agents.
Services is the new software (Pat/Julian). Examples:
- Medicine: inspect genome, personalize recs, prescribe, clinical trials.
- Law: negotiate contracts, litigate, settle.
- Math/sciences: solve AIME, discover superconductors.
- Consumer: manage inbox/calendar/finances/taxes.
Agents everywhere b/c easier than humans: scalable w/ compute, low maintenance, cheaper (tokens < salary). Humans smarter now, but bitter lesson—soon smarter at many things. Humans adapt.
Deployment swift/unprecedented: economics clear, scalable.
Agents ballooning exponentially/super-exponentially—things get strange: agent commerce (pay/negotiate?), swarms policing (cybersecurity/megadeath?). World weird fast.
Long-horizon agents here, clear curve.
Founders: compressed timelines (Nathan/Zed: 3yr moonshot over holidays w/ Claude Code; Brett Taylor: Sierra weekend; Notion: 8M lines code in 6wks). Stacking them = 100yrs progress in 100 days.
Bifurcate work: physical (package Pony Express/satellite Falcon 9; force x distance) vs. cognitive (Pythagoras theorem/DeepMind protein folding).
Similar revolution pattern.
Physical Work Revolution
Most history: muscle (people/animals). ~1700+: water/wind, steam, combustion, electric motors. Today (2026): 99+% physical work by machines (plane here, goods mfg, transport).
Cognitive Work Revolution
Most history: humans (+ sliver mechanical: astrolabe/clock). Recent: electronic computation (trillions calcs now serving humans). Neural nets next—soon 99.9% cognition by machines.
Cognitive revolution like industrial: bigger/faster.
-
1800s: Washington Monument—tallest building, capped w/ 100oz most precious metal (aluminum), displayed Tiffany's. Decades later: electrolysis → cheap (candy wrappers, trash). Aluminum = intelligence; AI = electrolysis. PhD skills invoked instantly, then discarded.
-
Alien design: World optimized for human cognition. Machines: different. 2006 NASA antenna: traditional symmetric → evolutionary algo/RL: alien shape, dramatically better. AI designs (chips/cars/buildings) not intuitive to humans—be open-minded.
It's going to have alien design. The third motivation story is on emerging sciences. In the early industrial revolution, great engineers like Newcomen and Watt perfected combustion engines: put petrochemical into a piston, ignite on fire, millions or billions of particles explode, move the piston, work. For almost a hundred years, it was tinkering—engineers improving it incrementally, perhaps seeing scaling laws, playing with the product.
Over 120 years after Sadi Carnot came around and formalized this in a new science: thermodynamics. He said, "Wait a second. There are millions or billions of particles. We can actually formalize what that all looks like."
In AI, there are billions of neurons, trillions of tokens. Right now, we're in the tinkering phase. Even if we think it's an understood science, it's not. In the future, we will have a science as fundamental as thermodynamics introduced in the next couple decades. Someone in this room might come up with that science. It will be taught in high schools. It will be that fundamental. It will help us master AI and even master consciousness.
For the vast majority of human history, tens of thousands of years, art has been a progression towards realism: cave painting from about 25,000 years ago, Egyptian hieroglyphs, Greek pottery, Renaissance paintings—a grand transformation toward realistic art.
Then engineering came along: Daguerreotype, early photography. What was spent decades of life perfecting—getting every brush stroke perfect—gone. The world thought painting was over: "The machine can do it better than any human. Art is ended."
Humans responded: Was the purpose of art to capture the moment the way the eye sees it, or the way the heart and soul sees it? This led to impressionism, expressionism, cubism, neoexpressionism—new forms of art as humanity's response to this dramatic change in science.
2500 years ago, Greek philosopher Protagoras wrote, "Man is the measure of all things." Nothing in a vacuum has value to humans—not aluminum, not art, not intelligence. It only has value because of the experience.
AI can do the work. AI will do the work. But only the human connection can give you a reason to care. That's why we're all in this room today.
A decade from now, work is going to be dramatically different. Things are going to change so much. But the one thing that will be constant is the relationships you form today with the person right next to you—they will endure. That's what you're going to look back on. That's what's going to be valuable from today. Form those relationships with the people next to you, enjoy your time together at this AI Ascent, and really lean into what makes us most human.
Keep Sequoia Capital in your library
Save the videos and channels worth coming back to, and find them again in one place.
