The web's bot-versus-bot future

The big picture: As AI agents improve and multiply, bots representing individuals will interact with bots representing companies, and human use of the open web will continue to decline.

• My bot will talk to your bot — but you and I will probably talk a lot less.

Zoom out: In the early '90s, the World Wide Web introduced a people-to-people network.

• As websites got more elaborate and apps proliferated, the network evolved into a people-to-machines space, as our interactions with other people became heavily mediated by software (think: social feeds or travel apps).
• Next up: an online world where interactions take place primarily among bots and AI "agents," with people largely relegated to the sidelines.

Zoom in: Take one of the most basic things we do today — buying stuff online.

• We're used to a world in which you click around, check products and prices from different vendors — maybe with help from a comparison service or website. While prices can fluctuate and algorithms sometimes play a role, as on Amazon or Uber, the purchase decision remains firmly in human hands.
• But AI-driven e-commerce means that vendors are going to start rapidly changing their prices based on your identity and other variables — not, just say, once a day but by the microsecond, and differently for each customer.
• Delta Airlines is already experimenting with one AI ticket-pricing tool.

Yes, but: These pricing systems can rapidly become complex "beyond human cognitive limits," as one study has already found.

• To cope with that, AI makers will respond by offering buyers their own AI-based tool to represent their interests in transactions with AI pricing bots.
• Then we will all sit back and watch as AIs representing both sellers and buyers have at it.

This kind of spiraling digital arms race is most familiar today in the realms of electronic trading and cybersecurity, where offense and defense have long played a "see if you can top this" game.

• The same brutal competitive dynamics are about to spread everywhere — to job applications and classrooms, dating apps and customer service, coding helpers and scientific research.

By the numbers: SEO experts already believe that half of the visitors to websites today are bots rather than people.

• ChatGPT is on track to be talking to "billions of people a day," per OpenAI CEO Sam Altman, and at some point the chatbot will "say more words a day than all humans say."

Here are some signposts to watch for over the next months and years that will confirm bot hegemony has arrived.

• The replacement of traditional SEO with "GEO" — generative engine optimization — could accelerate, with products and content owners no longer competing for top placement in Google but instead seeking to become chatbots' first "answer." One top GEO technique involves using AI to write AI-friendly pages — bots writing for bots.
• Trading bots trained to maximize returns will start to figure out — as Bloomberg's Matt Levine has speculated and at least one study has predicted — that their best strategy is to collude in manipulating the market.
• Ransomware gangs are already deploying chatbots to negotiate with their victims, as Axios has reported. Next up: Targets of ransomware attacks will let AI agents handle their response to the attacks — leaving all the people involved free to relax on the beach, assuming there's any money left in their bank accounts.

In this new world, a simple website or app will no longer provide adequate digital representation for companies, organizations and individuals.

• Service providers and large organizations are already preparing to roll out two different versions of each website or app they support: one for people and the other for AI agents and bots.
• "We invested all of this energy into optimizing websites for human user experience, and now there are all of these nonhuman users who have an entirely different set of needs," Linda Tong — CEO of Webflow, the site-building service provider — told Axios.

AI bot activity — scrapers compiling data to train models and agents seeking to provide answers to human users' questions — is already beginning to swamp traditional websites.

• Cloudflare, the website service provider, recently accused AI search engine Perplexity of ignoring site owners' efforts to block scraper bots. Perplexity said the bots Cloudflare objected to were actually AI agents helping individual users.
• As such disputes spread, open source developers are creating new tools for websites to deploy in an effort to block aggressive AI bots — another arms race that's likely to intensify.

Between the lines: The more bot interactions take over online spaces, the harder it will become to monitor and understand what's actually happening on any particular website or inside a company.

• We will need more bots just to keep tabs on what all those bots are doing.
• "In a few years, you're going to have hundreds or thousands of agents on your network that are all interacting with employees or interacting with each other," Jeremy Burton, CEO of startup Observe, told TechCrunch last month. "That's all great until something goes wrong, and you've got to try and, you know, do a Sherlock Holmes and figure out who done it."
• Crucially, the U.S. legal system has barely begun to figure out how to apportion blame when an AI agent goes awry. AI makers will claim that liability lies with the user who instructed the agent, while users will blame the agent's provider.

Our thought bubble: The high likelihood that a world of bots will fuel arms races everywhere could foil the prime intent of AI makers to make our lives more efficient.

• As bots compete to over-optimize the realms they operate in, they may end up introducing extravagantly wasteful new kinds of automated friction — a flood of transactions and interactions that don't actually accomplish anything or improve outcomes for either side of the deal, but do cost a fortune in software and energy costs.
• The willingness of investors and VCs to write humongous blank checks for money-losing AI development gives tech firms an incentive to build this glum world even if the people who ultimately drive markets don't want it.

What's next: In a bot-dominated world, the only solution that seems to make any sense is more bots.

• "By training AIs to fight and defeat other AIs we can perhaps preserve a healthy balance in the new ecosystem," science-fiction author Neal Stephenson suggested in a recent talk.
• Stephenson proposes we design AIs "to predate upon existing AI models by using every conceivable strategy to feed bogus data into them, interrupt their power supplies, discourage investors, and otherwise interfere with their operations."

This is the first of a two-part series. In part two, humanity strikes back.