The a16z Show
Technology S-Curves and the Waves That Follow
Benedict Evans maps the arc of technology through S-curves — PC, smartphones, mobile — and identifies four emerging waves poised to reshape everything: autonomy, mixed reality, crypto, and AI. He argues these are not separate stories but cascading platforms, each at a different construction stage, with AI already in its rocket-fuel phase and machine learning becoming a universal enabling layer like relational databases.
The talk itself: youtube.com/watch?v=cVYDkPidXrU
Twenty-four minutes for the cleanest version of the argument that technology waves look obvious afterwards and invisible during. It was easy to predict mass car ownership; nobody predicted Walmart.
The argument, in 9 sections
13 quotes, none repeated01S-Curves Stack on Top of Each Other
Every major technology platform follows an S-curve: slow start, explosive growth, maturation, then boredom. The PC Internet was followed by the mobile Internet, and as each curve flattens, attention shifts from the platform itself to what gets built on top of it. On PCs we built search and social; on smartphones we built ride-sharing, Instagram, Instacart. When a platform matures, people declare innovation dead — but that's always wrong.
Evans points to Marc Andreessen arriving in Silicon Valley in 1994 thinking he'd missed everything — only for the web to explode immediately after. The pattern repeats: there's always something else coming.
“Innovation is dead and this is really boring and the new iPhone looks just like the last iPhone and what have you done for me lately. And of course it always feels like that.”
The declaration that innovation is over is itself a recurring pattern — every mature S-curve produces this complaint right before the next one arrives.
“A young student called Marc Andreessen came out here and said he'd already missed the whole thing, that everything had already happened. And of course that wasn't quite right.”
The feeling of having missed the boat is perennial. What follows is usually the biggest wave yet.
02GAFA: Super-Evolved Organisms That Still Look Vulnerable
Google, Apple, Facebook, and Amazon operate at unprecedented scale — Amazon heading toward 600,000 employees, Google and Facebook capturing half of all internet advertising revenue. They're founder-controlled, they read the disruption playbook, and they compete directly against each other. Yet every previous dominant company looked invulnerable until it didn't.
The scale difference matters: Microsoft had 45,000 employees in 2000, while today's giants employ 5-10x that. But size alone isn't immunity. Microsoft's total dominance of computing was rendered irrelevant by the iPhone. Something will always come along.
“The winners always look invulnerable until they don't. So IBM looked invulnerable, then Microsoft, Intel, Nokia did, then the internet giants of the last bubble. Now Google, Apple, Facebook, Amazon look invulnerable — but all previous ones fell, and it was not predictable at the time how or why they would fall, but they all did.”
03The Four Emerging S-Curves, Each at Different Construction Stages
Evans frames the next waves using a skyscraper construction metaphor:
Autonomy: Still “down in the muddy hole digging foundations” — the technology doesn't work yet but will.
Mixed Reality: Just emerging from the foundation phase to where you can “throw the frame up” — comparable to multi-touch in 2006.
Crypto: The frame has gone up and “we're just trying to work out what the facade will look like and who the tenants will be” — technology works but product-market fit is unclear.
AI: “The tower is up and we're filling it up with tenants” — deep into the rocket-fuel phase.
S-curves are never smooth. The iPod “took almost five years to work — overnight success that took five years.” Each technology goes through three phases: the tech doesn't work but you believe it will; it sort of works but no product-market fit; then it explodes.
“Overnight success that took five years.”
S-curves are deceptive. The iPod and iPhone both looked like sudden explosions, but both went through years of the technology kind-of-working without product-market fit.
04Machine Learning Is an Enabling Layer, Not a Product
The best analogy for machine learning is relational databases in the 1970s and '80s. Databases went from record-keeping (replacing paper filing cabinets) to business intelligence systems, enabling entirely new billion-dollar companies — Oracle, SAP, and eventually every enterprise software company (Salesforce, SuccessFactors). But nobody looks at Starbucks and calls it “a database company” — even though it's enabled by databases.
Machine learning finds patterns in data — cats in photos, fraudulent behavior, delivery routes, customer churn. The key insight: “is there a cat in this picture” becomes the same kind of question as “which customers are going to churn” or “is this transaction fraudulent.” Pattern recognition is a generalizable solution.
“Nobody looks at Starbucks and says this is a database company, but they're all enabled by databases. That I think is the best way to think about machine learning.”
Machine learning should be thought of the same way — not as a standalone product, but as an enabling layer that transforms every industry.
05Automation Gives You Washing Machines, Not Humanoid Robots
Every wave of automation gets imagined as general-purpose AI — 1950s domestic robots walking around the kitchen — but what we actually get is specialized machines: washing machines, dishwashers, microwaves. Evans calls these “the washing machines of machine learning.”
The data and solutions are extremely specific to each vertical. You don't get one general-purpose AI — you get hundreds of narrow applications. The multiplying effect is what matters: “a steam engine doesn't replace one horse, it gives you a thousand horses. Machine learning doesn't replace one person — it gives you the capability to do what you could only have done if you had a thousand of those people before.”
“Automation and imagine it's intelligent, but of course Charlie Chaplin falls into the machine — it doesn't know it's Charlie Chaplin, it just carries on doing the thing that the machine has been built to do.”
06Computers Learning to See Will Reshape Commerce, Cities, and Industry
For 50-75 years, computers could count and read text but couldn't process images or video. Machine learning changes that. Evans argues this alone would make ML “the biggest thing in the tech industry.”
Applications cascade outward: show your phone a bag and ask what it is and where to buy it; point at a plant and diagnose disease; put cameras over every step of an iPhone assembly line and analyze the industrial process; count everyone in a subway station or shopping mall automatically. The image sensor becomes a universal input.
“You turn the image sensor into a universal input for a computer — it's not just a computer that can see like people, it's computers that can see like computers, and that makes all sorts of things possible.”
07"Driverless Car" Is Like "Horseless Carriage" — The Real Change Is Everything Around It
Calling an autonomous vehicle a “driverless car” makes the same conceptual error as calling early automobiles “horseless carriages” — it preserves the old form while missing the transformation.
Once you remove the steering wheel and combustion engine, you can completely reimagine the vehicle. Cars can platoon at 150 mph inches apart. Roads, lanes, and signaling get redesigned around computers that “don't drive like people.” Then cities transform: no accidents, no parking constraints, dynamic real-time road pricing — changing where you live, shop, eat, and drink.
“Saying a driverless car is like saying horseless carriage. They're taking the horse off but they haven't actually changed anything else. And of course that isn't how cars evolved.”
The real transformation of autonomy isn't the vehicle — it's roads, cities, parking, retail, and daily life rebuilt around the new capability.
“It was easy to predict massive car ownership but hard to predict Walmart.”
The first-order effect of a new platform is visible; the second-order economic and social reorganization is the real upheaval and is nearly impossible to foresee.
08Mixed Reality at 2006 Multi-Touch Stage; Crypto at 1994 Web Stage
Mixed reality is where multi-touch was in 2006 — prototypes work, the interaction model is taking shape, everyone in the room screams when they see the demo, but the shipping product is still ahead. It enables not just placing objects into the world (Minecraft on a table, recipes hovering at the stove) but the computer looking at what you're looking at: recognizing faces, products, remembering where you left things.
Crypto is where the web was in 1994: the technology works fine, but “it wasn't quite clear what you were going to do with it and how useful it would be and what it would mean if everybody on earth was using this.” If the internet automated retrieval of information, crypto automates trust. Money went to the cloud centuries ago — into banks, then paper, then databases — but was always inert. Crypto makes those records active, programmable.
“That database can do stuff and those records in that database can do things and mean things that weren't really possible with any previous way of storing this kind of information.”
09Visible Activity ≠ Real Progress
The construction site metaphor: you walk past a hole in the ground for six months, see nothing happening, then one Monday the frame is up. Another six months of apparent inactivity, then suddenly the skyscraper is finished.
This maps to the S-curves: autonomy is still in the hole, mixed reality is emerging to frame-up stage, crypto has the frame and needs tenants, and AI's tower is already being filled. The lesson: the periods that look boring or stagnant are often when the most important foundational work is happening.
“There's very visible activity that doesn't necessarily correlate with the period when actually stuff is being done or important things are being created.”
Also worth keeping
2 more · each links to its minute“The winners always look invulnerable until they don't.”
GAFA's unprecedented scale doesn't exempt them from the pattern — every dominant tech company in history has eventually been disrupted.
“A steam engine doesn't replace one horse, it gives you a thousand horses. Machine learning doesn't replace one person — it gives you the capability to do what you could only have done if you had a thousand of those people before.”
Reframes automation from subtraction to multiplication — the point isn't fewer humans but vastly more capability.
What to take away
5 lines- 01Frame AI as infrastructure, not product
Treat machine learning like databases — an enabling layer that every company will use without being “an AI company.” The question isn't “what's our AI strategy?” but “what becomes possible now that finding patterns in data is a generalizable solution?”
- 02Ignore the “innovation is dead” narrative
It's a predictable symptom of a maturing S-curve. While the conversation fixates on the old platform's stagnation, the next wave is already being built — likely invisible in its foundation-digging phase.
- 03Beware “horseless carriage” thinking
When evaluating new technologies, don't just imagine removing the old constraint (horse, driver, central authority). The real change comes from what gets redesigned around the new capability — roads, cities, trust models, business structures.
- 04Watch the construction stage, not the visible activity
Autonomy, mixed reality, crypto, and AI are at fundamentally different stages of readiness. The periods of apparent stagnation are when foundational work happens. The frame goes up overnight — but only because of months of invisible foundation work.
- 05Machine learning's biggest unlock may be computer vision
For 50+ years computers could count and read but not see. Turning every camera into a universal input — able to recognize objects, count people, analyze industrial processes, spot trends in millions of photos — may be the single most transformative application, independent of AGI timelines. ---