AI Just Started Building Interfaces on the Fly. Your Website Is Still Static.
Joey Alvandi·March 16, 2026
Anthropic shipped a feature last week where Claude builds a custom calculator when you ask about compound interest. Ask about pasta carbonara, and you get a recipe card with adjustable servings and a "Get cooking" button. Ask a comparison question, and you get an interactive table instead of a wall of text.
The AI reads what you're trying to do and generates a purpose-built interface — in real time, from scratch, unique to your question.
This is available to every Claude user, including free accounts. OpenAI launched something similar the same week. Google shipped interactive charts for Gemini Ultra back in December.
Three companies spending billions on UX research all arrived at the same conclusion within 90 days: when someone is trying to understand something or make a decision, the interface should be built for them in the moment — not pre-designed for the average case.
Here's the question that's been stuck in my head since: if AI can generate a custom interface for a simple cooking question, why is your B2B website — with its complex product, multiple buyer personas, and high-stakes purchasing decisions — still serving the same static pages to every visitor?
The Principle Behind the Feature
This isn't really about Claude, or recipes, or calculators.
Anthropic's design team looked at a conversation about dinner and concluded that the right response wasn't a paragraph of instructions. It was a purpose-built interface with interactive elements, generated in the moment, based on what the user actually needed. They could've just listed ingredients. Instead, they built something the user could work with.
That's generative UI — the AI decides what format will be most useful, then builds that format on the spot. Different user, different question, different interface. No one pre-designed it. No one maintained a template library for it. The AI generated it because the context demanded it.
Now think about the last time a prospect landed on your pricing page. They got a static table, maybe three columns of feature comparisons, and a "Talk to Sales" button at the bottom. They had to mentally map their own use case to your packaging, compare tiers themselves, and figure out which features mattered for their specific situation. All from a layout designed identically for every visitor who's ever loaded that page.
Anthropic decided that wasn't good enough for a recipe. Your complex B2B product deserves at least the same standard.
We Measured This Gap Across 516 Websites. It's Worse Than You Might Think.
I'm not guessing about how badly B2B websites handle this. We have the data.
In Q1 2026, we audited 516 B2B websites through NavigationAudit.com and will be publishing the results this week in The State of B2B Website Navigation. 82.4% of websites earned an F for navigation quality. The average visitor clicked through 14 pages and spent 7.4 minutes searching — and still, fewer than 35% of their questions got answered.
Only 5% of sites earned an A or B. You can run the same audit on your own site at NavigationAudit.com — takes about two minutes and scores your navigation friction on the same scale we used across the full study.
We also found something counterintuitive: sites organized around buyer type actually scored worse on our friction scale than those organized by product taxonomy (5.1 vs. 7.5). The best-performing industry — Sales & Marketing software — answered 67% of buyer questions. The overall average across all industries was 35%.
Even the winners left a third of buyer questions on the table. That's the gap generative UI closes — not by redesigning the layout, but by generating the right experience for each visitor in the moment.
Why Did Chatbots Fail Where Generative UI Succeeds?
I can already hear the objection: "We tried AI on our website. We have a chatbot."
We tested those too. Our State of B2B Chatbots report (Q4 2025) audited 100 mid-market B2B SaaS chatbots with four questions any real buyer would ask: "What is [Company]?" "What makes you different?" "Show me a case study." And a use-case-specific question.
66% couldn't answer "What is your company?" The most basic question a visitor can ask. 83% failed when asked for a case study. Of 34 companies that labeled their chatbot "AI-powered," only 9 passed all four questions.
The problem wasn't the technology. It was the architecture. Chatbots were built to qualify leads — to collect an email address before delivering any value, to route you to a sales rep, to interrupt your browsing and ask you to do the work of explaining what you need.
Generative UI works in the opposite direction. It doesn't ask for anything upfront. It reads context, performs tasks for the user — building the right interface, assembling relevant information, creating interactive tools — and presents the results without demanding anything in return. That's the fundamental architectural difference.
The Macro Shift: Why This Is Bigger Than One Feature
According to Gartner, machine customers — AI agents that autonomously research, evaluate, and purchase on behalf of human decision-makers — will influence $30 trillion in spending by 2030.
Gartner separately forecasts that by 2028, 90% of B2B buying will be intermediated by AI agents, pushing over $15 trillion through automated exchanges.
The interfaces people use to research, evaluate, and buy things are being fundamentally rebuilt. The pre-designed, static layout — the model that's dominated web design since the late '90s — is losing ground to generated experiences that adapt to what each person needs in the moment.
Consumer AI tools are leading this shift. Hundreds of millions of people now interact daily with interfaces that respond to their context. Then those same people land on a B2B website and get a mega-menu with 47 links and a hamburger icon on mobile.
The expectation gap grows every time Anthropic, OpenAI, or Google ships an update. The same buyer who just used Claude to generate a custom comparison table will land on your site and get a static PDF download behind a form. That's a trust problem, not a design problem.
What Should Marketing Teams Do Next?
I've been sitting with this the past week, and here's where I've landed.
First: stop treating your next website redesign as the solution. Redesigns take 6–12 months and produce another static layout that starts aging the day it launches. The pattern Anthropic validated isn't "better static design." It's generative UI — experiences built dynamically for each visitor. You can't redesign your way to that.
Second: evaluate any on-site AI against the "does it perform tasks for the buyer" test. If the first thing your AI widget does is ask for an email or company name, it's a qualifier. That's the old architecture. The new standard is technology that reads context and generates relevant pathways — assessments, solution maps, personalized recommendations — without demanding information upfront.
Third: start making your content machine-readable. Gartner's $15 trillion projection isn't about human buyers clicking through your navigation. It's about AI agents consuming structured data through APIs and generating their own buyer experiences. If your content isn't structured for machines to parse, you're invisible to the next generation of purchasing processes.
Where Navless Fits in This Picture
I want to be transparent about my perspective.
I'm the Director of Marketing at Navless. We've been building for exactly this shift — and the Claude announcement is the clearest validation yet of the principle our product is built on: if AI can generate a custom interface for a question, it should do it on your website too.
Guide sits on top of existing websites as an agentic layer. Under the site, AI agents do the work — figuring out what each buyer needs and building the right screens for them.
The buyer sees a self-guided discovery experience: a choose-your-own-adventure journey where every screen is tailored to their role, their problem, and their buying stage.
That's generative UI applied to B2B websites. Deploys through a single embed code. No site redesign required.
The same principle Anthropic's design team just validated at the platform level.
My Personal Assessment
Here's what I keep coming back to. Anthropic, OpenAI, and Google didn't coordinate this. They independently converged on the same principle: when someone is trying to make a decision, the interface should be generated for their specific context — not pre-designed for the average case.
That principle has been true on B2B websites for years. We've just been papering over it with mega-menus, search bars, and chatbots that ask for your email before they help you. The data from our research — 82.4% of sites earning an F, 83% of chatbots failing basic questions — shows how wide the gap is.
If AI can build a custom calculator for a compound interest question, it can build a custom experience for a buyer evaluating your product. The only question is how long your website waits to catch up.
How is your website handling this shift? Still static pages, or something that actually builds itself around the person in front of it?