Summary

Automated traffic already surpassed human traffic in Cloudflare’s data in May — years ahead of the team’s forecasts — and, with the caveat that his prior predictions have been wrong, Prince says that if current trends continue, agentic traffic could reach “a thousand times as much as human traffic in five years.” He guessed H2 2027 at Web Summit, revised to H1 2027 at South by Southwest in March 2026, then his team reported it had already happened. The kicker for every ad-driven business: “bots don’t click on ads,” so “the entire business model of the internet is about to change radically.”
Cloudflare is hunting for a micropayments rail that could scale three orders of magnitude beyond Visa — and is open to crypto or other partners and acquisitions. Roughly half a billion web transactions/sec flow through Cloudflare; 1–10% monetizable implies ~10M financial tx/sec, scaling to 100M “in very very short order,” versus Visa’s 80k peak and ~2M equities/sec on the busiest market day ever. The best vendors seen reach ~2M tps via blockchain hybrids — “still an order of magnitude slower than we need on day one.” NET Dollar is “an idea of a stablecoin… we have a name,” announced to flush partners out of the grass.
The layoff thesis behind Cloudflare’s roughly 20% RIF (Jason cited about 1,000–1,100 people): AI can take on more measuring work — audits, finance, legal, compliance and middle management — while builders and sellers remain hiring priorities. AI’s biases are “uncorrelated with the rest of the organization,” giving it a fresh set of eyes. Manager spans moved from roughly 6 toward 12 direct reports, flattening the org. Peers say they need to act but won’t — “that’s just chicken leadership”; every day of waiting is “a disservice to the employees that you’re eventually going to lay off.”
Prince wants Google held to the same data rules as other labs: “they shouldn’t be able to leverage their monopoly of yesterday in order to get monopoly tomorrow.” Google sees 2x the internet OpenAI does and 4x Anthropic; Prince says every other AI lab exists in part because of an “almost religious opposition” to Google’s attitude — back to the reportedly described Sam/Elon/Larry Page dinner where Page supposedly called the humans’ defenders “speciesist.” About 80% of major AI companies use Cloudflare.
Agent commerce could threaten the ad complex: Prince estimates average US consumers may spend ~$1,200/year on agents versus Google’s ~$750 and Meta’s ~$250 per US user — a prize that ad-based models may struggle to win because “ads inherently are like you’re corrupting the system.” If your agent recommends a Ford because Ford paid, you switch instantly. His unsolved worry: the ad-funded web serves the “kid in Sri Lanka” for free; a pay-to-play agent internet may not.
Architecture edge: containers can’t power the agent era — one agent per knowledge worker would need “40 times the number of CPUs that are produced in the world.” Cloudflare’s isolates, based on browser-tab technology, make it a preferred deploy target for Lovable, Replit, Base44 and Codex-generated code; more than 20% of the internet and almost half the Fortune 500 already sit behind it.
The unpriced infrastructure risk nobody discusses: during COVID a mere 2x traffic spike nearly collapsed parts of the internet, forcing a “Mexican standoff” where Netflix, Amazon and YouTube all cut 4K streaming. “That was at 2x. We think it’s going to be a 1000x” — without a working payment model, “there’s a lot of stuff that’s just going to break.”

Deep dive

1. Bots passed humans earlier than even Prince’s team predicted

Prince’s team kept revising its forecast: asked at Web Summit in Lisbon when automated traffic would pass human traffic, its data said the second half of 2027; the same question at South by Southwest in March 2026 produced the first half of 2027; then in May the team reported, “we’ve actually already passed automated traffic over human traffic.”
With “the big caveat that I’ve been wrong in every prediction so far,” Prince says that if current trends continue, agentic traffic could reach 1,000x human traffic in five years, not because humans use the internet less but because bots grow so fast that “humans are really going to just almost be a rounding error online.”
The economic break: ads have been “the only micropayment system that has worked at real scale” — but “bots don’t click on ads. And so the entire business model of the internet is about to change radically.” What replaces it is, in Prince’s words, “the most interesting question in the world,” with deep implications for media, polarization and how society is organized.

2. “Everything wrong with our world today is Google’s fault” — the attention machine

Prince first credits Google as “the hero of the last 27, 28 years” — not just search but Analytics and the DoubleClick infrastructure; “without that we really wouldn’t have the web in the form that it is.”
Then the line he admits gets him in trouble every time: “at some level everything that’s wrong with our world today is Google’s fault.” The attention-monetization model begot TikTok and media engineered to trigger “either dopamine or cortisol” — making people irrationally happy or irrationally angry so an ad can be sold against it.
His optimism: the next business model could “reward actual knowledge creation” instead of engineering fight-or-flight responses — “I’m pretty optimistic about what the positive impacts on society could be if we get that right.”

3. Cloudflare as anti-Facebook, and the web is growing again for the first time since 2012

When an early investor asked who Cloudflare’s biggest competitor was, Prince said “Facebook” — because when running your own site meant security and performance pain, people gave up and launched Facebook pages instead; in Nigeria “the internet is Facebook largely.” His framing: “Cloudflare is to Facebook as Shopify is to Amazon.”
From roughly 2012 through early 2025, web growth was minuscule and at times negative — more websites were taken offline than created. Late 2025 changed that: AI let people who never considered themselves developers ship sites and applications, producing growth that “looks closer to what the internet growth looked like in the early 2000s.”

4. Every AI lab is partly a religious opposition to Google — Prince wants the field leveled

OpenAI’s origin, as Prince recounts a reportedly described story: it followed a dinner conversation between Sam, Elon and Larry Page, where Page supposedly said that if AI out-evolved humans, humans deserved to be extinct; Sam and Elon defended humans, and Page reportedly called them “speciesist.” “Every other AI lab that exists exists in part because they started as almost a religious opposition” to that attitude inside Google — which had DeepMind’s researchers, its own silicon through TPUs, and the most data.
The asymmetry Prince keeps citing: Google sees twice as much of the internet as OpenAI, four times more than Anthropic, and many times more than X or Meta. Jason also raised Prince’s reported push to unbundle search from AI crawling. Prince’s stated goal is that Google should not be able to “leverage their monopoly of yesterday in order to get monopoly tomorrow.” He wants “not one or five… 500,000 amazing AI companies,” and says about 80% of major AI companies use Cloudflare, including companies among the Chinese labs.
Jason’s probe on a possible contradiction — Prince signed the letter against regulating open-weight models while demanding labs pay for training content. Prince sees no conflict: open weights and creator compensation can coexist; without cost, you get a tragedy of the commons — a free-to-crawl internet means a bot asked “where should I go to lunch” will “literally go and scour every page on the internet and then come back and say you should go to Wendy’s.”

5. State of the union — and why isolates, not containers, power the agent era

The scope: a network in more than 350 cities within milliseconds of most of the world’s population; more than 20% of the internet behind Cloudflare; almost half the Fortune 500; almost all major crypto companies; and since 2017 a developer platform that has become a preferred deployment target when Lovable, Replit, Base44 or OpenAI’s Codex generate code.
The container math is the load-bearing argument: if every knowledge worker runs just one agent in a container — each importing a full OS and toolchain — “you need 40 times the number of CPUs that are produced in the world.” Isolates, the technology behind browser tabs, are far lighter and faster; Prince credits “somewhat dumb luck” for having built the right architecture.
On their own adoption, disarming honesty: Cloudflare once pitched itself as an AI company and “got so laughed out of that room”; they were “actually slow adopters” of this wave — paranoid about shipping internal data to unknown vendors — until mid-2025 made the impact “blindingly obvious,” and late October–November 2025 (Jason: right as Opus 4.6 came out) forced restructuring of the organization itself, not the products.

6. Builders, sellers, measurers — and the case against “chicken leadership”

The Wall Street Journal op-ed taxonomy Jason recapped: builders make things, sellers sell them, and measurers handle audits, finance, legal, compliance and middle management. Prince’s underappreciated insight is that AI has biases too, “but those biases are uncorrelated with the rest of the organization” — a fresh set of eyes, versus teams’ dangerously shared biases.
The mechanics: most salespeople went from spending maybe 40% of their time with customers to about 90%; manager spans moved from the Harvard-recommended roughly 6 direct reports toward 12 — and since average span is the simplest measure of hierarchy, the org flattened and middle management shrank. Meanwhile, if developers become 10 times as productive, Prince says he will hire as many developers as possible.
The cost was real — “really dear friends of mine that we let go,” death threats, a stock dip that “went right back up” — and the placement effort, severance and continued vesting were deliberately unusual. His indictment of peers who say “we got to do that too” but won’t move: “that’s just chicken leadership… once you realize you need to make these changes, every day you wait is actually a disservice to the employees that you’re eventually going to lay off.”

7. More than 1,000 interns and the manager-to-IC reversal

Zag when everyone zigs: Cloudflare plans to hire more than 1,000 interns over 2026 — Prince initially said 1,111, then restated the total as “over 1,000” — because “they’re all AI native” and everyone else stopped. Half the deal is training them; the other half — “you have to go train us” — embeds interns in senior teams to push new tools, with knowledge transfer running both directions.
The shape of work is inverting: senior people, including people going to places like Anthropic, are asking to return to individual-contributor roles. Jason’s Blockworks anecdote lands the point — a team lead refusing another human hire: “I’ve got eight agents right now… I hate human management, but I’ve never had more fun managing these agents.”
Prince explicitly rejects the doom camp of “Dario and some of the people in the labs”: builders and sellers thrive (“if you’re doing a $100 million software deal, that’s still going to be humans”), but “if you’re getting a degree right now in accounting, I’d be a little bit more nervous.”

8. Why Prince prefers public markets to venture: no-fault divorce

The analogy he leads with: spousal homicide rates are lower in jurisdictions with no-fault divorce — and private markets have no no-fault divorce. You can’t fire your VCs and they can’t easily fire you, so “nobody tells the truth.” Public investors just call: “That was stupid. I sold your stock and here’s why” — and half the time he agrees.
Against the short-termism cliché: Scottish investor Baillie Gifford’s average holding period is 16 years, versus VC funds that last 10 years, 12 at the absolute maximum; on Cloudflare earnings calls “our CFO doesn’t get very many questions — everyone is focused on what’s the strategy.”
Most misunderstood aspects: customers still see the “security CDN” — now a small slice of the business — while investors who call it SMB-heavy miss that SMB is under 10% of revenue, with most coming from large enterprises.

9. The 10-million-transactions-per-second problem nobody has solved

Prince deflates the grand-strategy reading of pay-per-crawl, the wallet and NET Dollar with a Dark Knight reference — “everyone thinks they have a plan” — but the direction is clear: flip the tragedy of the commons so agents pay “tiny fractions of fractions” per resource, funding both infrastructure and the creators whose content powers the bots.
The math that disqualifies everything existing: ~half a billion web transactions/sec through Cloudflare, 1–10% monetizable → roughly 10M financial tx/sec, probably scaling to 100M. Visa peaks around 80k tps during holidays; the busiest day in financial-market history saw ~2M equities transactions/sec. The best vendors and potential partners reach ~2M tps with blockchain-based hybrids using batching and pooling — “still an order of magnitude slower than we need on day one,” and two orders soon.
NET Dollar is candidly “an idea of a stablecoin… we have a name” — announced as “flushing the pheasants out of the grass” to surface partners. Cloudflare is building and shopping in parallel, “hiring like crazy” for the internal team, with “no pride of ownership” — partner or acquire. Jason floats the first big token M&A deal involving a beaten-down L1 with real technology; Prince says, “I would not be surprised if there’s some team that’s thinking about that.”
Jason’s 2018 Blockworks blog predicted browser crypto wallets for web-wide micropayments — a reminder, both agree, that “things tend to take much longer than you think they will.”

10. Not religious about blockchains — and the perils of touching money

Why touch a chain at all rather than internal ledgers? Chains work across organizations; the problem is “consensus is slow” — so Prince guesses the answer “ends up being something that’s a little bit hybrid.” His anti-dogma line: “we got a hammer, but we got a screwdriver and we’ve got a drill” — funds move in and out via crypto rails and PayPal, Visa, Amex and banks alike, with Cloudflare as “the universal translator,” transacting between sites and agents without holding the money.
Jason’s pushback — worth keeping: Cloudflare’s business today is neutral plumbing; money is regulated, custodial and potentially adversarial in new ways. Prince notes he is already an unwilling expert in “Spanish copyright law” and “Indonesian politics,” and says regulators have been constructive because they too fear the web’s business model imploding. “I’m not naive… we will have to deal with all sorts of new problems,” fraud at micro-scale included.
Italy’s Piracy Shield is his line in the sand: anyone running a media company can list a site for takedown globally, with no judicial oversight — they even tried to take offline a site created by the US State Department. A court order to block in Italy, he’d obey; but “an unelected cabal with no judicial oversight can block something everywhere in the world — obviously we’re going to push back on that because that’s insane.” The dark joke: an Italian white-collar conviction lets you serve time in-country — “I can just go hang out in Tuscany for a while.”

11. What hasn’t become obvious yet: 2x nearly broke the internet, and 1000x is coming

The buried COVID story: traffic doubled in two weeks and parts of the internet almost collapsed. Prince was on a “Mexican standoff” call with Netflix, Amazon, YouTube and Akamai as Asian and European peak hours aligned across European exchange points; everyone eventually agreed to cut 4K streaming to keep the network up. “That was at 2x. We think it’s going to be a 1000x.”
The new-entrant problem he can’t solve: startups win customers through convenience or emotion, but agents are “coolly calculating and they have almost infinite amount of time” — and will favor things they know and can find the most information about, compounding consolidation. He wants a world of new entrants, “but I don’t know that we know exactly how that’s going to work.”
The $1,200 fight: Prince guesses he may personally spend ~$1M/year on agents; the average US consumer ~$1,200, potentially replacing Netflix-type entertainment spend, versus Google’s ~$750 and Meta’s ~$250 per US user. Ad-based agent models could struggle because “ads inherently are like you’re corrupting the system” — if the agent pushes a Ford over a Nissan because Ford paid, users defect. The casualty he keeps returning to: the ad-funded web serves the “kid in Sri Lanka” for free — “I haven’t heard any solution to that problem,” and it risks locking in developed-world advantages.
The interface shift and the fame economy: browsing on desktop, social and mobile all retained advertising as the dominant business model; the shift to an AI-chat interface will put enormous pressure on it. The 14-year-old whose internet is TikTok, Prince’s 85-year-old father living in chat, and George Jetson asking Rosie illustrate the transition. From his Stockholm conversation with Spotify’s Daniel Ek: people create to get rich or famous, and fame motivates more (“totally irrational — being anonymous and rich is the best thing in the world”). Since LLMs are “a mathematical model of all of human knowledge,” Prince proposes a Nobel Prize or Academy Awards measuring who contributed most to human knowledge each year — because paying creators isn’t enough if the AI CliffsNotes version never makes them famous.

Dino Mavrookas: America Needs to Build Ships Again Inside the Factory Building America’s Autonomous Navy