Software, not a person, loading your client’s website: reading it, clicking through it, and acting on it on someone else’s behalf.
Also known as agentic browsers, AI browsing, autonomous web navigation
Agentic browsing is an AI agent operating a browser the way a person would: navigating to pages, reading what is there, filling in fields, pressing buttons, and moving between sites until a goal is met. Find three quotes for commercial insurance and summarize the differences is a browsing task, not a search query, and it ends with the agent having actually visited the sites.
It differs from crawling in both intent and behavior. A crawler indexes for later; an agent is trying to finish something for a specific person right now, and it will interact with the page to do it. That makes it a visitor with very particular habits. It reads text and structure, it does not linger, it is largely indifferent to motion and layout, and it can be stopped cold by obstacles a human steps past without noticing.
Once a meaningful share of a client’s traffic is software completing a task, a lot of craft assumptions get quietly tested.
Design decisions that assume a viewer break. Text baked into an image, meaning carried by layout alone, a consent wall, a full-screen interstitial, an email gate: each is a small friction for a person and a hard stop for an agent.
Analytics gets murkier. Agent sessions can register as traffic without behaving like it: no scroll depth, no dwell, an immediate exit once the one needed fact is extracted. Read without adjusting, that pattern looks like a content problem and gets treated as one.
The brand’s writing becomes its interface. An agent summarizing a client for its user is working from the page’s actual words and structure. Clear product names, plain specs, and consistent terminology now carry weight that visual identity used to carry alone.
A direct-to-consumer client launches a site that wins a design award, and conversion on one channel drops anyway. The hero is a full-bleed video with the offer typeset inside it, and the size guide lives behind a hover state. Shoppers using an assistant to compare three brands get a summary of the client that mentions neither the offer nor the fit information, because from the agent’s side of the glass, that content is not on the page. The remedy is not a redesign. It is making the same claims available as real text, giving specs proper markup, and having somebody check what an agent actually reports back before the site ships.
The automations and agents module of the workshop teaches you how to build AI workflows that compress the busywork without taking the craft out of the studio.