Your Next Customer Won't Have Eyes: Why Tieyiwe Bassole Says We Must Design for AI Agents

Since last year I have been making an argument that sounded strange in meetings: we should start designing products for AI agents, not only for humans. This autumn the data caught up with the argument. Automated traffic already makes up more of the web than human traffic, the part made by AI agents is growing faster than anything else online, and agents now have a common language, the Model Context Protocol, for using our software. I believe the companies that redesign for this first will win the next decade, and the ones that keep designing only for eyes and thumbs will slowly disappear from view.

The numbers

Several independent measurements now point the same way:

• Imperva, which publishes an annual report on bot traffic, found that automated traffic passed human activity for the first time in a decade in 2024, at 51% of all web traffic. In its 2026 report it put the 2025 figure above 53%, with human activity down to 47%.
• Cloudflare, which carries a large share of the world's websites, said in March that bots generate about 57% of the requests to sites on its network. Its chief executive, Matthew Prince, said he had expected that crossover around 2027, and that the growth of AI agents brought it forward. In October Cloudflare reported that traffic from AI agents had grown by more than 1,700% in a year.
• HUMAN Security found that automated traffic grew eight times faster than human traffic over the past year, and that traffic from agent tools rose by several thousand per cent.

These are estimates, and each company measures its own slice of the internet. Not every bot is an AI agent, and not every bot is bad: a lot of this traffic is search engines and AI features working on someone's behalf. But the direction is not in doubt. My own forecast is that within six months to a year, AI agents on their own, not just bots in general, will be the largest group of visitors to many business websites. When that happens, the visitor you most need to serve will not have eyes.

Software is becoming de facto open source

The second half of my argument is about what agents do with software. An AI agent can read your documentation, study your interface, and rebuild a working copy of much of your product. The developer Martin Alderson reported cloning most of the core of Linear, a popular project-management tool, in about 20 prompts with an AI coding agent, over a couple of evenings. A team on the Latent Space podcast replaced a tool costing $43,000 a year with coding agents, and found that copying the software was the easy part; the hard parts were taste, integration and maintenance.

That is what I mean by de facto open source. Your code does not have to be published for its ideas to be freely reproduced. When anyone with an agent can rebuild your features, features stop being the moat. What remains valuable is what an agent cannot copy in an evening: your data, your relationships, your trust and track record, the integrations you maintain, and how easy you are for other agents to work with.

Agents already have a language for our products

The Model Context Protocol, MCP, is the standard way for an AI agent to discover and use a tool. There are now more than 10,000 public MCP servers by Anthropic's count, and the protocol's software kits are downloaded about 97 million times a month. According to reports, in July the maintainers finalised the biggest revision of the specification so far.

The web is adapting too. Google is testing WebMCP in Chrome, a proposed standard that lets a website declare its functions as tools an in-browser agent can call directly, instead of making the agent guess which button to press. Cloudflare launched "Markdown for Agents", which serves a clean text version of a page to agents that ask for it. Not everyone agrees on the details: Google's John Mueller called converting pages to Markdown for bots "a stupid idea", and one analysis of about 300,000 domains found only one in ten had adopted the llms.txt guide file, with no clear sign it helps. But the question is no longer whether agents will use our products. It is whether our products will make sense to them.

What designing for agents means in practice

Designing for agents does not mean abandoning people. Humans still decide, approve and pay. It means changing who you design the first draft for. These are the rules I now apply:

1. Every important action gets a machine door. If a person can book, buy, cancel or ask for a quote, an agent should be able to do it through an API or an MCP server, with clear names and inputs.
2. Pages must make sense without seeing them. Real headings, labelled forms, buttons that say what they do, content that does not depend on hovering or animation. This helps screen readers too.
3. Publish facts as data. Prices, stock, opening hours, delivery times and policies in structured form, so an agent does not have to guess from a design.
4. Give agents their own permissions. Scoped access, spending limits, and a human approval step for money, deletion and anything irreversible.
5. Write errors for machines. "Something went wrong" helps no one. "The delivery date must be within 30 days" lets an agent fix the request.
6. Measure agent visits separately. If agents are half your traffic, they deserve their own funnel, their own analytics and their own conversion rate.
7. Write documentation once, for both. Plain, structured, versioned text that an agent can read and a person can skim.
8. Keep the human moments human. Trust, taste, care and the final yes still belong to people. Design those parts for people, and everything else for both.

This is the direction I am pushing our own work at TIBLOGICS. The businesses that make themselves easy for agents to understand will be the ones agents recommend, choose and buy from. In my comparison of Grok Bot, Muse and Dots, you can see how capable these agents already are. They are coming to your website. Make sure it speaks their language.

Sources

• Imperva: Bad Bot Report 2026, bots in the agentic age
• Thales/Imperva: 2025 Bad Bot Report, AI-driven bots surpass human traffic
• CNBC: AI and bots have officially taken over the internet, report finds
• NBC News: Bot web traffic has overtaken human web traffic
• Forbes: Cloudflare says AI agent traffic grew more than 1,700% in the past year
• TechCrunch: Online bot traffic will exceed human traffic by 2027, Cloudflare CEO says
• Martin Alderson: Attack of the SaaS clones
• Wikipedia: Model Context Protocol
• InfoQ: WebMCP standard proposal available in Chrome origin trials
• InfoQ: Cloudflare debuts Markdown for Agents
• Digiday: WTF is Markdown for AI agents?