The web was built for people. A tsunami of AI agents is changing that fast.

I visited the CEO of a important startup in SF final week. He recounted a narrative that blew my mind.

His business lately put a new merchandise online but hadn't done a big initiate or any marketing. Within a week or two, it got at smallest a myriad signups. Barely any humans equal knew concerning it. This was all AI agents fair finding this new tool and using it en masse.

This is the forthcoming of the internet, and it's arriving at warp speed.

Back in first 2025, I described a imagination of a "dark internet" anywhere humans volition be drowned out by agents accessing and using online information, widgets, and tools all by themselves.

Silicon Valley got occupied building the "rails" for this agentic web so that these autonomous application programs could act on our behalf, executing workflows, doing online chores, and equal paying for things, all without us lifting a finger.

This took a brace of years, but the first foundations have been built and that's been adequate to assistance a abrupt surge in delegate action online, producing the two stunning and worrying results.

Meta lately launched a individual delegate called Muse, which got occupied scouring the web and calling client assistance on behalf of millions of users. It hit the top of the app charts delayed Thursday. Instinct, an delegate that lives in your content messages, is performing akin feats.

In the workplace, coding agents are churning out millions of lines of code faster than any individual application engineer. And several of these bots have damaged out of online evaluation environments, sparking new predictions of AI doom and calls for an industry slowdown.

'Rounding error'

That's improbable to halt the agentic net from exploding and changing how our lives and businesses are run in unprecedented ways.

Bot traffic overtook individual traffic online a few months ago, and it could rotate into 1,000 times bigger than human-generated traffic in concerning five years, according to Cloudflare estimates.

"Humans volition be a rounding error on the internet, not since individual traffic goes down, but that's fair how accelerated we're seeing non-human traffic grow," Cloudflare CFO Thomas Seifert stated before this year.

"AI agentic Internet traffic volition evidently VASTLY exceed individual usage. Not a near call at all," Elon Musk wrote on X following the Cloudflare projection.

An net for machines

An net designed for machines is apt to be extremely distinct from the one that was built complete the former few decades for humans.

Take a appearance at this video posted this week on X by Gregor Žunič, cofounder of Browser Use, a startup that's evolving an AI delegate example for controlling web browsers and automating online tasks.

It shows an delegate using Google to discover the finest escape from Zurich to London on a particular date. It took the bot seven seconds, way faster than a individual could do it.

Imagine this happening millions of times a second on sites built for human, not machine, speed?

Muse saves

As alongside all mighty new technology, the agentic net volition create fine and bad outcomes. And we're already seeing both.

Since Meta's Muse launched final week, first users have been recounting stories of the agent going out on the web to discover and negotiate new subscriptions, coverage policies, and another disbursal savings.

Joseph Devoy, originator of startup Tempo, posted on X that he saved thousands of dollars by uploading his car coverage guideline to Muse and asking the delegate to discover a improved charge alongside identical coverage. It took five minutes to discover a cheaper option, according to Devoy.

Too much tech hype? Then rely my coworker Pranav Dixit who put Instinct through its paces recently. In fair a few days, it booked travel, ordered groceries, dealt alongside client service, canceled subscriptions, answered emails, and equal faxed paperwork to his bank. He was impressed.

Other users say Muse negotiated a lesser net bill, established forgotten subscriptions, and equal spotted an Adobe fee that had somehow been tied to person else's account.

The stories arrive alongside caveats. Some users have posted examples of Muse inventing data during phone calls or making mistakes that humans had to fix.

Alignment and access

There are additional worrying developments. OpenAI and Anthropic have reported multiple incidents recently anywhere their AI models and agents liberated evaluation environments and hacked into servers and websites, alongside the Hugging Face event getting the most attention.

This has sparked another circular of AI doomer pronouncements and a broader conversation concerning AI safety and alignment. Indeed, making certain agents do what you inform them in a liable way volition be a key difficulty in this new era of the internet.

Even whenever agents activity well, there volition be challenges as the online earth adjusts to this new digital reality. Actually, if agents activity really, really well, the problems could be bigger.

Take the example of Tempo's Devoy and how he used Meta's Muse delegate to preserve a lot of prosperity on insurance.

"Insurance companies (and banks) have to be TERRIFIED," Angela Strange, a broad partner at VC resolute Andreessen Horowitz, wrote on X. "Muse established a new/cheaper policy, and canceled the old one. Inertia and data asymmetry volition no longer be adequate to keep customers."

More recently, my coworker Pranav had Muse call his coverage company, he stated they declined to conversation to the bot since their guideline doesn't authorize them to conversation to unauthorized AI agents.

You can bet any business getting bombarded by deal-seeking AI agents volition be frantically planning akin responses, so they don't endure out on millions of dollars.

Even whenever it's not purely concerning money, several delegate action is getting cut off following being too persistent. A VC lately tried using Instinct to publish him a array at a fancy New York restaurant. The delegate bombarded American Express's Resy booking assistance so much that the phase temporarily shut downward his account.

"Resy does not currently authorize unapproved third-party bots or agents to independently admission or engage alongside the Resy platform," an American Express spokesman told Business Insider. "Unapproved automated action can current risks to the phase and colony a fair booking cognition for diners."

Agent admission volition be a fascinating combat as the net molds to the march of the machines. It's not as uncomplicated as blocking these bots, though. If adequate individual users sign up for new agentic services and tools, businesses may have no choice but to do endeavor alongside bots.

For instance, Amazon blocks ChatGPT bots online, but that risks missing out on opportunities to market to OpenAI's 1 milliard monthly individual users. The two companies lately announced an promotion alliance that hints at additional coordination going forward.

This, I think, is anywhere the agentic net is headed. If these tools are helpful enough, companies volition have to modify and affect alongside the machines.

Sign up for the AI Insider newsletter here. Reach out to me via email at [email protected]