Deep dive
The Dead Internet Theory Is Half Right
Keith Brown · Founding Partner, Outperform Ventures
Nov 8, 2026 · 8 min read
The dead internet theory is right about work and wrong about people. AI agents are taking over the online chores, and the time we save goes to entertainment.
I went to get Starbucks this week and took this photo. Every person at the tables was looking at a phone.

I kept staring at YouTube's numbers this fall. Engagement keeps going up while almost everything else people do online is getting automated. I started asking what happens when people are removed from the digital work entirely.
My answer is a version of dead internet theory for business. The work part of the internet goes quiet, handled by AI agents talking to other software. What people keep doing online is personal: watching, connecting and spending their own time.
What is the dead internet theory?
Dead internet theory is the idea that most of what's online is now made by bots and AI rather than people. It started as a fringe theory, but the traffic numbers have moved toward it. Automated traffic made up more than 53% of all web traffic in 2025, up from 51% the year before, according to Imperva's 2026 Bad Bot Report. Human traffic fell to 47%.
I think the theory is half right. It's right about business, where agents are taking over the work. It's wrong about personal life, where people are online more than ever to be entertained.
i never took the dead internet theory that seriously but it seems like there are really a lot of LLM-run twitter accounts now
— Sam Altman (@sama) View on X
Imperva's report calls AI agents "a new category of internet participant." They retrieve data, run workflows and act on behalf of users. That last part is where business comes in.
Cloudflare, which carries a large share of the world's web traffic, sees the same thing. This year, for the first time, more than half of internet traffic it handled wasn't human, Cloudflare wrote in November. Daily requests from AI agents on its network grew by more than 1,700% over the past year. Its network went from 63 million requests a second at the end of 2024 to 115 million today.
Welp, that happened faster than I predicted. Thought it would be end of 2027, then early 2027, but agentic traffic growing so fast that bots have now passed human traffic online for the first time in the Internet's history.
— Matthew Prince, Cloudflare CEO (@eastdakota) View on X
What do people do online today?
Walk through a normal week. Each of these is a task an AI agent can handle or soon will:
- Search Google for an answer
- Buy something on Amazon
- Pay a bill
- Plan a trip and book the flight
None of these is something people do for fun. They're chores done through a browser. Once an agent can do them, there's little reason for a person to sit at the screen.
I saw this up close this week. I submitted an MCP server to Anthropic's directory, and automated checks approved it before I could ask how long review would take. I wrote about it in Anthropic approved our MCP server without a human.
What happens when half your visitors are software?
Cloudflare's data shows what this does to businesses online. In retail, software, IT and financial services, human traffic fell as much as 40% in less than a year. AI answer engines read the page and give the person a summary, so the visit never happens.
Training bots are a growing share of the machines. They made up 22% of AI crawler requests in spring 2025 and 52% by June 2026, by the crawlers' own stated purpose. Fewer than 1% of sites on Cloudflare block search crawlers, while 17% block AI training.
Cloudflare draws a line between those crawlers and agents doing a job for a person:
An agent booking a table, comparing insurance quotes, or buying a dataset for a researcher is a customer. It just isn't a human one.
Websites are starting to charge those customers directly. Cloudflare's Monetization Gateway, in closed beta, lets a site put a price on each request an agent makes.

Thousands of sellers joined the waitlist, and their most common request was "charge agents, not humans." It's the same change I saw from the builder's side with our MCP server: agents calling tools instead of people clicking pages.
We'll Only Go Online to Be Entertained
So what will we all do online? Mostly be entertained. I don't think screen time goes down. Once agents handle the work part, we won't have a reason to be online to work, and the hours shift from business screen time to personal screen time. That's my prediction, and YouTube's numbers point the same way.
YouTube took 15.8% of all US TV viewing time in August 2026, its highest share ever, per Nielsen's The Gauge. That's double Netflix's 7.7%, Tubefilter reported, and it happened in a month when total TV use fell 4.4%.
It was 7.9% in February 2023, the first month Nielsen counted YouTube on its own, according to Mark Studios' tracking of Nielsen data. YouTube's share doubled in about three and a half years.
Viewing among adults 50 and older rose 9% in that report. Older viewers are watching more YouTube on their TVs, including podcasts. To me that fits the pattern: as the online chores shrink, the time goes to watching and listening.
The internet becomes a place for leisure
Today's internet is a mix of work and pleasure. I think the work part goes away almost completely. What stays is content, connecting with people and personal time.
| The internet today | The internet I expect | |
|---|---|---|
| Search and research | People typing queries | Agents asking other software |
| Shopping and bills | People clicking through checkouts | Agents ordering and paying |
| Support and sign-ups | Forms and queues | Automated checks and approvals |
| Watching and connecting | People | People |
That's my prediction, not a measured fact. The Imperva, Cloudflare and Nielsen numbers are the measured part, and both are moving in the direction I'm describing.
Is this what the SaaSpocalypse is about?
People call what this year has done to software businesses the SaaSpocalypse. Most of the talk is about AI writing code and companies building their own tools. I think the bigger story is people leaving digital work altogether.
A lot of software is priced per seat, one seat for each person who logs in. If agents do the logging in, the seat count falls. I covered what replaces that model in software dollars moving from SaaS to custom.
Correct take. 99% of tokens in the world will get consumed in an enterprise context. They will be for writing code, combing through insane amounts of life sciences research, automating manufacturing, securing enterprises, detecting fraud, and thousands of other tasks like this that have significant economic value. This is both where the cost of AI can be most justified but also where the impact of parallel workers doing tasks on your behalf is most useful. Even the agents that most impact consumers will often be wrapped up as just end to end services where the consumer doesn’t think about there being an AI system on the other end, like faster insurance or onboarding to a bank. The key though is it’s going to take years for diffusion of agents across the economy because workflows have to re-engineered to incorporate AI, which is a complete change to how most work happens today. Anyone who’s expecting this to happen overnight should be updating their timelines.
— Aaron Levie, Box CEO (@levie) View on X
Employers expect machines to do most of the work on 34% of tasks by 2030, the World Economic Forum found, up from 22% in 2025. That's the conservative view. The bull case is bigger. Today's technology could already automate about 57% of US work hours, according to the McKinsey Global Institute. McKinsey says that measures what the tools can do, not a forecast of job losses, because adoption takes time.
Physical work comes later. Robots are moving into warehouses and factories, but they're years behind what software agents can already do online.
People go back to the real world
Screen time moving to entertainment doesn't mean all of it. With no reason to be online for work, some of those hours go offline too. I think we're headed for a renaissance of the real world: more time with family, outside, in person with customers and teams.
That's also why I look for companies with a physical connection at Outperform Ventures. I wrote about the reasons in physical moats over SaaS.
A note to the leaders reading this
I've spent almost 20 years building and backing software companies. For most of that time, the job was getting a person to visit a page, sign up and log in every day. That's changing faster than anything I've seen in my career.
When I look at a business now, I start with the chores. The forms, renewals, bills and support tickets your customers put up with are the first work an agent will take from them. I'd make it easy for an agent to do that work with you, because if you don't, it will do the work with someone else.
Then I look at what's left when the chores are gone. People will still watch, connect and show up in person. The businesses I'm most excited about at Outperform Ventures are the ones customers would choose to spend that time with.

I don't know how fast this happens. I do know I watched an approval process run from start to finish without me this month, and YouTube's share of TV went up again the same quarter.
Related reading
Follow what I'm building and backing
I post on LinkedIn and X about the companies I back and the products I build.
Join 2,310 people following along.