The future of web: stop asking what to build, start asking what to build for

I wanted to question a sentiment I keep hearing from clients whenever we discuss AI and agents. It usually arrives as a version of the same question: what will the world look like in three to five years, and how do we prepare for it? How do we plan for a future web?

There's a lingering anxiety underpinning that question, and as an AI specialist and growth marketer with limited coding knowledge and no developer skills, it's one I wanted to investigate honestly. I've made my peace with not knowing what the future holds, and I hope you can too, because there's a clear trajectory we can embrace and prepare for instead.

If you'd asked me five years ago, before the arrival of GenAI, I'd have given you a confident roadmap that turned out mostly to be wrong. If you'd asked me six months ago, my caveat-laden prediction would already need thorough revising. Anyone selling you a five-year web blueprint right now is selling you certainty nobody has. The only thing I'm confident about is that I can't see past the next six-month development cycle.

That sounds like an unsatisfying hedge, but it isn't. While I can't tell you what the web actually looks like in 2031, I can identify the trajectory it's already on. AI is exploding, customer behaviour is following, and the businesses that win are the ones building to iterate at speed. You don't need to predict the destination now, you just need to stay within one cycle of the next wave so you're ready to ride the next one.

And once you stop asking what to build, a better question comes into focus. Not what should the website be, but how is technology reshaping the way our customers find, evaluate and buy. How do we meet that journey at the right moment?

The journey is compacting, and getting darker

Start with what customers are actually doing, because that's the part you can measure.

Buyers used to come to your website early and browse what's available. Now most of the journey happens before they ever visit your site, fill in a form, or show up in your analytics. Industry research puts 70 to 80 percent of the modern B2B buying journey in this invisible space: AI chatbots, peer forums, independent research, LLM queries. No cookies, no attribution, no trace. By the time someone reaches out, 83 percent of buyers have already fully defined their needs and 92 percent start with at least one vendor in mind.

So the journey is compacting. Research that used to spread across a dozen of your pages now happens in a single AI conversation, and it's going dark because that conversation isn't on your site.

This is where the anxiety usually kicks in. People look at the dark funnel and assume AI is stealing their sales, when in fact the opposite is happening. New technology isn't taking the buyer away from you; it's giving them an easier and more valuable path to purchase, and your job is to meet them where the new technology takes them. Just because people aren't clicking doesn't mean they aren't buying. The best brands will embrace that shift and insert themselves into the more convenient path, rather than focusing on how they expect the customer to behave.

From storefront to assistant to agent

The clearest way I've found to explain where this goes is to think about three eras of what a website actually is.

The web we built was a storefront, a window display with finite shelf space. You chose what to put in the window, your flagship products, your key applications, the current campaign, but most of what you sell never reached the shelf because there's finite real estate. The buyer saw what you decided to show them, not necessarily what they needed in that visit. This is the website most companies still have today.

The web your customers now expect is an assistant, and the shop becomes a query. The whole inventory of everything you've ever made is readily available, not just what fits in the window, as long as it's structured, described and findable. The customer asks a question, and the site answers it with seamless navigation to what they need. The constraint shifts from real estate to data architecture. If your content is structured and machine-readable, an assistant can surface it; if it isn't, it might as well not exist.

The web that's coming is agent-to-agent. The customer doesn't come to your warehouse personally, their agent does. A procurement platform acting on the buyer's behalf queries your systems directly, compares pricing and availability, negotiates terms, and returns a shortlist of three fully scoped specifications at machine speed. The buyer reviews three options that are ready to act on. They never see the suppliers with systems that couldn't respond.

This isn't speculative, and the infrastructure that makes it work is being laid right now:

Adobe shipped an MCP server for its commerce platform in March 2026, the first time an AI agent could browse, configure and check out on a major B2B platform with no human involved.

Stripe, Tempo and Visa put a machine payments protocol into production, removing the last human step so agents can initiate, authorise and settle B2B payments on their own.

MCP and A2A, the protocols that let agents talk to tools and to each other, became standards over 2025 and are now backed by the Linux Foundation, with well over a hundred organisations already in production.

Procurement-automation start-ups built specifically to take human browsing out of B2B buying are raising serious money, because the demand is coming from buyers, not from tech companies.

Three eras, three modes of consumption: humans to engage, assistants to inform and agents to transact. Right now you're mostly serving the first; the second is arriving fast, and the third is closer than the timeline suggests.

Build for humans first, then assistants, then agents

Here's where the customer-centric approach turns into something you can act on. If those are your three audiences, you build for them in order, because each one is the foundation for the next.

Humans come first, and they always will. The work here is familiar, and that's the point, because it's the work that makes everything after it possible:

Intuitive navigation and a clear path from problem to solution

Fast, mobile-ready, accessible pages

Content organised around what the customer is trying to solve, not the products you happen to sell

Assistants come next. An AI assistant doesn't browse your site; it reads the underlying code, so it needs:

Structured, answer-first content with clean semantic markup

Schema it can parse, attribute and cite

Full catalogue coverage, because an assistant can only recommend what it can read, and most B2B businesses publish full content on only a fraction of their SKUs while the rest sits in internal systems, invisible to buyers and AI

Agents come last, and they raise the bar again. An agent doesn't want a page; it wants an endpoint:

Programmatic access to specs, pricing and availability at machine speed

API endpoints and an MCP server it can query directly

An A2A-compatible layer so it can coordinate with other agents

Think of it like offering a USB cable to your digital storefront. The customer's agent accesses the cable, plugs it in and can programmatically query your whole offering machine-to-machine.

The reassuring part, and Google said this plainly in its own developer guidance, is that everything you do to make a site ready for agents also makes it better for humans. You're not building three websites. You're building one in an order that compounds.

This is also what makes the volume manageable. Full coverage, in multiple languages, structured for machines as well as people, sounds impossible if you picture a team writing it all by hand. The same AI shift reshaping the buyer's journey is what lets you produce and maintain content and infrastructure at a scale that wasn't realistic two years ago. Content, language and infrastructure are about to boom, and smart, AI-assisted platform development is how you keep pace with market demand.

Three underlying pillars to futureproof your website

When a client asks me how to organise all of this, I give them three pillars and one foundation, as a way of thinking rather than a rigid roadmap:

Platform: the architecture that keeps every other option open. Composable, semantic, agent-readable, built to change because the destination will.

Discoverability: being found by humans, assistants and agents alike, which now means earning citations and focusing on brand visibility, not just rankings.

Experience: helping every visitor, whether a person or a machine, find the answer and act on it. Everything from chatbots and smart search to full product coverage, and linguistic, topical and application depth.

Underneath all three sits the data foundation, because analytics, personalisation and agent-readiness all fail without a coherent data layer. None of it works in silos, so building a robust data warehouse will shine some light on the dark journey and provide strategic clarity to different business functions.

I'm not a developer. I came to this from a growth marketing perspective, and I can't write a lick of code that builds any of it. What I can do is watch the journey change and work out where it's heading, and the pattern holds: technology moves, customer behaviour follows, and the businesses that meet the new behaviour early are the ones that get cited, shortlisted and chosen.

Build for now, plan for the trajectory

So I've made my peace with not knowing what the web looks like in five years. It's the wrong thing to plan around.

Build for the customer you have now, the human who wants a fast, clear, helpful site. Plan for the customers arriving next, the assistants and agents acting on their behalf, and keep your architecture loose enough that you're never more than one development cycle from being ready for whatever comes after that.

You can't predict the next wave, but you can be in a position to ride it.

We work with B2B organisations to build websites that serve humans, assistants and agents in the same architecture, from structuring content for AI discovery to laying the technical foundations agents can query directly. If you're wondering whether your site is ready for the next wave of buyer behaviour, that's usually the right moment to talk.

Get in touch with the Comprend team →