Agentic Websites Are Here: Why Your Site May Soon Do Things, Not Just Say Things
For twenty years, a website's job has been to say things: here's what we do, here's what we charge, here's how to reach us. Increasingly, that's not enough. Design blogs and product teams - Figma among the loudest - are describing a shift towards "agentic" interfaces: software that doesn't just present information but acts on a visitor's behalf. Booking the appointment. Processing the return. Walking a shopper through a catalogue and narrowing it down to the three products that actually fit.
We want to be clear about what this piece is not. It isn't about getting your site found by ChatGPT or Google's AI Overviews - we've written about that elsewhere. This is about something adjacent but different: building agentic behaviour into your own site, so it can do work rather than just describe it. Here's what we're seeing, and what we think is worth taking seriously now versus later.
What "agentic" actually means for a website
Strip away the buzzword and an agentic website is one where a piece of software - usually an LLM wired up to your systems - can take a goal ("I need to reschedule my delivery") and carry out the steps to achieve it, rather than handing the visitor a form and a phone number. That means:
- Reading context: order history, account status, stock levels, calendar availability.
- Making decisions within rules you've set (what can be refunded automatically, what needs a human).
- Taking action through your existing systems - booking engine, CRM, inventory, payment provider - via APIs.
- Reporting back in plain language, and escalating to a person when it hits the edge of its authority.
This is a step beyond chatbots. A chatbot answers questions about your returns policy. An agentic flow reads the order, checks it's eligible, generates the label, and confirms the refund - with a human only looped in if something's ambiguous.
Where this is already showing up
The pattern is clearest in a handful of everyday flows that are naturally rule-based and repetitive - exactly the kind of task that's tedious for humans and tractable for software:
- Booking and scheduling: appointments, tables, consultations, service call-outs, matched to real-time availability rather than a static calendar widget.
- Returns and support: eligibility checks, label generation, refund processing, and status updates without a ticket queue.
- Catalogue and product guidance: a conversational layer that asks a few questions and narrows hundreds of SKUs down to a shortlist, instead of relying on filters a visitor has to configure themselves.
- Account and billing tasks: plan changes, invoice retrieval, subscription pauses - the low-risk, high-frequency requests that currently clog up support inboxes.
Design teams at Figma and elsewhere are framing this as the next interface paradigm - less about pages and navigation, more about intent and outcome. Whether or not the terminology sticks, the underlying expectation is spreading: visitors increasingly assume a site should be able to do the thing, not just explain how to do it elsewhere.
The build isn't the hard part - the plumbing is
The conversational or interface layer of an agentic feature is, frankly, the easy bit now - the tooling for that has matured fast. What decides whether a project like this succeeds is the same thing that's always decided integration projects:
- Do your systems have usable APIs? Booking, stock, CRM and payment platforms vary hugely in how well they expose data and actions. Some of this work is discovery before it's development.
- Where are the guardrails? You need clear rules for what an agent can do autonomously versus what needs sign-off - a refund under £20 is different from a refund of £2,000.
- What happens when it's wrong? Every agentic flow needs a graceful fallback to a human, and a way to audit what the agent did and why.
- How is it tested? Unlike a form, an agentic flow has to be evaluated against edge cases, not just happy paths.
None of this is exotic, but it does mean an agentic feature is a proper project - closer to a systems integration than a design refresh.
Where we'd start
We wouldn't recommend rebuilding your whole site around this in one go. The pragmatic route is to pick one repetitive, rule-based task that currently costs you real staff time - a booking flow, a returns process, a product finder - and turn that single flow agentic first. Get the guardrails right, measure what it saves, then expand.
Practically, that means: audit which of your existing systems already have APIs; pick the highest-friction task as your pilot; define explicit boundaries for autonomous action; and build in a visible, easy escalation to a human throughout. Done well, it's not a gimmick - it's a website that actually does some of the work your team is doing manually today.
Keep reading
More from the studio.
Safeguarding and Compliance Online: What Grassroots Clubs Must Publish on Their Website in 2026
National governing bodies (FA, England Netball, British Cycling and others) increasingly expect clubs to publish safeguarding policies, welfare officer contacts
How We Built The Bridge of Lights: Designing a Memorial Platform
The design and sensitivities behind The Bridge of Lights - a memorial platform built from real grief, now used by 500+ families worldwide.
AI Website Builders vs a Real Design Agency: What Should a Small Business Actually Choose in 2026?
Wix, Framer and Squarespace are all pushing AI site generators hard, prompting the 'why pay a designer when AI can build my site' question agencies now field fr