For the last twenty years, businesses built websites for humans.
Then we learned how to make those websites readable by search engines.
More recently, we started thinking about how businesses appear inside ChatGPT, Perplexity, Claude and Google's AI answers.
But I think there is another shift coming after that — and it may be considerably bigger.
Your next customer might not visit your website at all.
Their AI agent could discover your company, evaluate what you offer, compare you with alternatives and decide whether you belong on the shortlist. Eventually, it may communicate directly with your company's own agent, request information, negotiate constraints and complete part of the transaction.
The human will still define the objective and approve important decisions. But much of the research and evaluation that happens before a purchase could move from human-to-business interaction to agent-to-agent interaction.
B2B is slowly becoming A2A.
AI is moving from answering questions to completing work
Until recently, the most important AI-distribution question was relatively straightforward: does ChatGPT mention my company when someone asks for a recommendation?
That question still matters. If an AI assistant consistently recommends your competitors and ignores you, you are losing visibility during an increasingly important stage of the buying journey. But visibility is only the beginning.
AI systems are moving from answering questions to carrying out tasks. They can research products, compare suppliers, inspect documentation, complete forms, communicate with other software and take actions across multiple systems.
At the same time, the infrastructure allowing agents to communicate is becoming more established. Google originally introduced the Agent2Agent protocol so independently built agents could discover one another's capabilities, exchange information and coordinate work. It was later donated to the Linux Foundation, becoming a vendor-neutral open standard. The protocol has now reached a stable specification and is supported across infrastructure from Google Cloud, AWS and Microsoft.
This does not mean every business will suddenly begin closing deals entirely between autonomous agents tomorrow. It does mean the foundations are being laid for agents to participate much more actively in commercial decisions.
The internet was originally built around humans navigating pages. The next version may increasingly involve software navigating capabilities.
What happens when your buyer delegates the research?
Imagine a founder needs an accounting platform for a new company.
Today, they might search Google, ask other founders, read comparison pages, visit five websites and book two demonstrations.
In the next version of that journey, the founder might tell an agent:
Find an accounting platform suitable for a UK software company with international contractors, Stripe revenue and fewer than ten employees. Compare the best options, identify any hidden costs and recommend the safest choice.
The agent then performs the research. It needs to understand which products serve that market, what they genuinely support, what they cost, how they integrate and whether their claims can be trusted.
It may examine the companies' websites, documentation, reviews, structured data, third-party mentions and technical interfaces. It might communicate with another agent representing the supplier to answer specific questions. The founder may only see the final shortlist.
At that point, the company's website is no longer just persuading a person. It is providing evidence to a machine acting on that person's behalf. That is a very different distribution environment.
Most companies are not ready to be evaluated by agents
Most websites are full of language designed to sound impressive to humans. They promise transformation, innovation, efficiency and growth. They hide important details behind forms. They spread product information across landing pages, help centres, PDFs and sales conversations. Sometimes the pricing page says one thing while the documentation implies another.
A human buyer can occasionally work around that ambiguity. They can interpret the marketing language, contact sales or make assumptions based on the brand. An agent needs something more precise: what the product does, who it is for, what it costs, which constraints apply and what evidence supports the company's claims.
This does not mean every website should become a database or every business needs to publish an A2A endpoint immediately. It means clarity is becoming technical infrastructure.
Clear positioning, structured information, consistent company details, accessible documentation and credible third-party evidence will not only help humans trust a business. They will help agents interpret it correctly. The companies that are easiest for agents to understand may become the companies that agents are most comfortable recommending.
AI visibility is not the destination
This is why I increasingly believe that AI visibility is an intermediate stage rather than the final category.
SEO helped search engines find and rank businesses. GEO helps generative systems understand, cite and recommend them. Agent readiness will determine whether AI systems can evaluate those businesses confidently and eventually interact with them.
These layers build on one another. An agent cannot choose a company it cannot discover. It cannot recommend a product it does not understand. It cannot trust a claim it cannot verify. And it cannot complete a task if the business exposes no reliable way to interact with it.
Getting cited in an AI answer is therefore valuable, but it is only the first sign that machines can understand your company. The larger opportunity is making the entire business legible to agents.
This changes what marketing means
Marketing has traditionally focused on influencing human attention. Companies compete for impressions, clicks, time on page and share of voice. They try to create a message memorable enough for someone to return later.
But agents do not have attention in the same way humans do. They do not choose a company because its landing page animation looks expensive. They do not become emotionally attached to a slogan after seeing it seven times. They do not necessarily reward the business producing the greatest volume of content. They evaluate whether a company satisfies the objective they have been given.
That pushes marketing closer to operational truth.
- Can the agent determine exactly what you sell?
- Can it distinguish you from competitors?
- Can it verify your claims?
- Can it understand your pricing and limitations?
- Can it access the information or functionality required to complete the task?
In an agent-mediated market, vague positioning does not merely reduce conversion. It may make the business impossible to evaluate.
Why this matters for Chad
I have been thinking about this a lot while building Chad. Today, a major part of Chad's work is helping businesses understand how search engines and AI systems discover, interpret and represent them.
But the direction is becoming clearer. The important question will not remain does ChatGPT mention my business? It will become: can an AI agent understand my business well enough to trust it, recommend it and eventually work with it?
That requires more than running an audit or adding schema to a website.
Different AI systems can inspect the same company and reach different conclusions. One model may identify a technical problem. Another may decide that the real issue is missing authority. Another may understand the company's core service incorrectly because the positioning is ambiguous. The founder is then left with multiple intelligent opinions and no clear operating decision.
The valuable layer is the system above those models: the one that separates objective evidence from interpretation, identifies agreement and disagreement, decides which problem matters most and maintains continuity as the business changes.
That is the direction we are building toward with Chad. Not another dashboard full of visibility scores. Not an endless queue of generic recommendations. A system that understands what is preventing a business from being discovered and trusted, determines the highest-value intervention and helps execute it.
The transition will happen gradually — and then feel obvious
Humans are not disappearing from buying decisions. Important purchases will still require judgement, trust and approval. Relationships will continue to matter. A founder will not blindly accept every recommendation made by an agent.
But humans do not need to disappear for distribution to change. If agents perform the first 80% of research, filtering and comparison, the businesses excluded during that process may never reach the human decision-maker. That is the part founders should pay attention to.
The first commercial internet rewarded companies that became accessible online. The search era rewarded companies that became discoverable. The generative-AI era is beginning to reward companies that can be understood and cited. The agentic era may reward companies that can be trusted, selected and acted upon by machines.
SEO was about helping search engines find you. GEO is about helping AI systems cite you. The next layer will be about helping agents trust you, choose you and work with you.
That is when B2B starts becoming A2A.
Also published on Medium.