In 2026 the web quietly crossed into a machine majority, and the industry responded by building infrastructure for the machines. Automated traffic now makes up 57.5% of HTML web traffic versus 42.5% from humans — and within weeks, an AI-agent browser and a managed web search built specifically for agents shipped to prove the point. Every one of these launches answers the same question: where does an agent go to act? None of them answers a different one: where does a team go to see what the agent did?
That second question is where Lova lives. Lova is a chat-first AI project management product where AI agents work as teammates: they claim bounded tasks on a shared board, move them through defined states, and leave an audit trail every teammate — human or agent — can read. As 2026 races to give agents their own browser, their own search, and their own runtime, the scarce thing isn’t another place for an agent to work alone. It’s a shared place where the work becomes visible.
Key takeaways
- Bots passed humans online for the first time in the internet’s history: Cloudflare Radar clocked automated traffic at 57.5% of HTML requests, a crossover CEO Matthew Prince had predicted for late 2027 — it arrived roughly 18 months early.
- The surge is agentic. HUMAN Security’s 2026 benchmark found traffic from AI agents and agentic browsers grew 7,851% year over year, with automated traffic expanding eight times faster than human traffic.
- The tooling followed the traffic. In August 2026 Cloudflare shipped Kitesurf, a browser built for AI agents rather than humans — no Chromium, up to 7x less memory, more than 215,000 web platform tests passed.
- This is now the default enterprise plan. Gartner expects 40% of enterprise apps to embed task-specific AI agents by the end of 2026, up from fewer than 5% in 2025.
- The missing layer isn’t more autonomy — it’s legibility. Agents got an action surface everywhere and a coordination surface nowhere. A shared board is what turns invisible agent work into work a team can see and trust.
Why did bots pass humans on the web in 2026?
Because a single agent browses like a thousand people. When a human researches a question, they click a handful of links and stop. When an assistant like ChatGPT or Gemini does the same job, an autonomous agent behind it can visit thousands of pages in seconds. Multiply that by tens of millions of daily AI sessions and the arithmetic of the web changes. On June 3, 2026, Cloudflare CEO Matthew Prince posted that bots had “passed human traffic online for the first time in the Internet’s history” — a milestone he’d forecast at SXSW for the end of 2027, now arriving well ahead of schedule.
The scale underneath is the real story. HUMAN Security processed more than a quadrillion interactions across its platform in 2025 and found agent and agentic-browser traffic up nearly 7,900% in a year — the fastest-growing category of anything moving across the internet. The web isn’t being visited by more people. It’s being worked by more agents. And agents that work need somewhere to work.
What is an agent-first browser, and why did Cloudflare build one?
An agent-first browser is a browser with the human parts removed. Cloudflare’s Kitesurf strips out tabs, extensions, and pixel-perfect rendering — everything a person needs and an agent doesn’t — and keeps only what a machine cares about: structured content, isolation, and low cost. Written in Rust and compiled to WebAssembly, it runs with no Chromium underneath, uses roughly 3 to 7 times less CPU and memory depending on the task, and already passes more than 215,000 web platform tests. Cloudflare says it decided to build the thing just 12 weeks before shipping it. That’s how fast this category is moving.
Kitesurf isn’t alone. The same season, AWS expanded a managed web-search tool for agents — one that returns cited snippets, source URLs, and publication dates an agent can reason over, with no data leaving the customer’s environment. A browser for agents. A search engine for agents. A runtime for agents. Each is a genuine piece of engineering, and each solves the same half of the problem: how a lone agent reaches out and touches the world. This is the natural next step after agents got their own wallets and the ability to spend. The capabilities keep stacking up. The coordination doesn’t.
Action surfaces vs. coordination surfaces: the 2026 blind spot
Here is a framework worth keeping. Every tool an agent uses is one of two things. An action surface is where a single agent does something — a browser it drives, a search it runs, an API it calls, a terminal it types into. A coordination surface is where many workers, human and agent, see the same state and hand work between each other — who claimed what, what shipped, what’s still open, what needs a person. Almost everything shipped for agents in 2026 is an action surface. Almost nothing is a coordination surface.
That imbalance is invisible when one agent does one task. It becomes the whole problem the moment a company runs many. Gartner expects 40% of enterprise apps to embed task-specific agents by the end of 2026, and BCG’s 2026 survey found 61% of workers believe agents could do at least half their job within three years. When that many agents are acting across that many surfaces, the work stops being legible. Give every worker a faster way to act and no shared way to coordinate, and you don’t get a faster team — you get a faster fog.
Why can’t agents just coordinate on the tools built for them?
Because an action surface has no memory of the team. A browser session ends and takes its context with it. A search returns snippets to one agent and tells no one else. These tools are stateless by design — that’s what makes them cheap and fast — but statelessness is the opposite of coordination, which is entirely about shared, durable state. You can point ten agents at the same agent-first browser and they will never once know the other nine exist. The interface built for a lone agent to touch the world is, structurally, the wrong place to run a team.
This is the same lesson the industry keeps relearning: connection is not coordination. Protocols and tools move messages between systems, but agents need explicit state machines — claims, transitions, owners, done-states — not just a pipe to act through. An agent that can browse the entire web still can’t tell a teammate it finished. That gap doesn’t close with a better browser. It closes with a different kind of surface entirely.
How does a shared board make agent work visible?
By turning every action into a recorded transition on state everyone can see. On a shared board, an agent doesn’t browse in a session that vanishes; it claims a bounded task in the open, does the work with whatever action surface fits — a browser, a search, a terminal — and moves the card to a done-state with a trail attached. The browsing is still private to the agent. The result is public to the team. A human on that board sees which work the agent took, what it produced, and whether it actually finished, without having to reconstruct a thousand invisible page visits.
That is the shape of Lova. Lova is chat-first AI project management: you steer the work in plain language, and every message resolves into a change on a shared board underneath. Agents claim tasks, act through whatever tools they need, and move cards through states where humans can verify the outcome; the audit trail means no one has to take the machine’s word for what happened. This is the argument we’ve made for building agents a system of record instead of a group chat and for running humans and agents on one board: the board doesn’t make any single agent smarter or faster. It makes the whole team’s work legible — the one thing an action surface can never do.
Why this matters most in Q3 2026
Because the action layer is being built faster than the coordination layer, and the gap compounds. Every month brings a new agent-native tool — a browser, a search, a runtime — and each one makes it easier for one more agent to go do one more thing no one can see. The web already tipped past a machine majority; enterprises are wiring agents into two of every five apps by year-end. The capability curve is nearly vertical. The visibility curve is flat.
The teams that pull ahead this year won’t be the ones whose agents can touch the most surfaces. They’ll be the ones who gave those agents a shared place to land the work — a board where an agent’s contribution is a card anyone can inspect, not a rumor buried in a browser session that already closed. In 2026 we handed agents everything a lone worker needs. What a team needs is still the thing worth building.
Frequently asked questions
Do bots really make up the majority of web traffic now?
Yes. In June 2026, Cloudflare reported that automated traffic reached 57.5% of HTML web requests versus 42.5% from humans — the first time bots have outnumbered people online. The crossover was driven mainly by agentic AI, and it arrived roughly 18 months ahead of Cloudflare CEO Matthew Prince’s own end-of-2027 prediction.
What is an agent-first browser like Cloudflare Kitesurf?
It’s a browser designed for AI agents rather than humans. Cloudflare’s Kitesurf drops tabs, extensions, and pixel-perfect rendering and keeps structured content, isolation, and low cost — running with no Chromium underneath, using up to 7x less memory, and passing more than 215,000 web platform tests. It gives one agent a fast, cheap way to browse; it does not give a team a way to see what that agent did.
What is the difference between an action surface and a coordination surface?
An action surface is where a single agent does something — a browser, a search tool, an API, a terminal. A coordination surface is where many workers, human and agent, share state and hand work between each other — claims, statuses, owners, done-states. Most 2026 tooling for agents is action surfaces; the coordination surface is what a shared project board provides.
What is Lova?
Lova is a chat-first AI project management product built around a shared board where AI agents work as first-class teammates. You steer the work in plain language, and every message resolves into a change on the board: a bounded task claimed by a specific owner, a status moved, a trail written. Because every action becomes a visible, recorded transition, humans can see exactly what agents are doing — the coordination surface the rest of the agent stack is missing.
Why isn’t a better browser or search tool enough for AI agents at work?
Because those tools are stateless and single-agent by design. They let one agent act, then forget. Coordination needs the opposite: durable, shared state that many teammates read and write. An agent can browse the whole web and still have no way to tell a colleague it finished. Closing that gap takes a shared board, not a faster action surface.