Topic
The agentic web
The way people find, evaluate and choose what they buy is being mediated by AI agents at every step. Brands that build their own agentic surface stay in the conversation. The rest get summarised away.
What is the agentic web?
The agentic web is the rewiring of how content and commerce are discovered, queried, and acted on by AI agents that increasingly mediate between people and the web.
Browsers are becoming destinations. Search is becoming conversation. Product discovery is moving from filter-and-browse to ask-and-be-answered.
This is not a future trend. It is underway in 2026. The brands that engage with the architecture deliberately stay in the conversation. The brands that wait get summarised away.
Why this matters now
Three forces are converging.
ChatGPT, Gemini, Perplexity and Claude are now where high-intent commercial queries begin. OpenAI announced product discovery inside ChatGPT in March 2026. Anthropic and OpenAI both ship AI browsers that act on the web on the user's behalf. The traffic that used to land on brand .coms is increasingly mediated by agents.
The data layer is becoming addressable. Digital Product Passports attach structured, verified data to physical products. The Model Context Protocol lets brands expose product data to LLMs as queryable sources. Both are infrastructure for agents to ground their answers in primary data.
The cost of being absent is rising. Brands without an agentic surface end up summarised by LLMs from whatever the model has scraped. Brands that engage with the architecture get to write what customers actually hear.
The three layers
The agentic web is layered. Brands engage with different layers, and the work of building a coherent surface across them is the discipline.
The surface layer
A brand-owned agent on the brand's own site, optimised for the customer's actual purpose: discovery, comparison, purchase, post-purchase support. Our product TalkPod is the working version of this at scale. Agentic commerce covers the storefront-specific slice.
The data layer
Digital Product Passports and machine-readable product information that the agent reasons over. Without it, even an agentic surface delivers patchy answers. See our Digital Product Passports page for the deeper read on the data layer.
The integration layer
The Model Context Protocol and similar standards that let LLMs query brand data directly, with citations the agent can use. Brands that engage here stay queryable by the LLMs customers are already asking.
The brands that win the next decade treat these as one architecture, not three vendor decisions.
Frequently asked questions
What is the agentic web?
The agentic web is the architecture for how content and commerce are discovered, queried, and acted on by AI agents that mediate between people and the web. It covers brand-owned conversational surfaces, integrations with LLMs and AI browsers, the structured data layer those agents reason over, and the content discipline that makes a brand citable in their answers.
How is this different from agentic commerce?
Agentic commerce is one application of the agentic web, the slice optimised for purchase. The agentic web is broader: discovery, evaluation, comparison, post-purchase support, and the data and integration layers that power all of them. We have a dedicated topic page on the commerce slice at /agentic-commerce/.
Why are AI browsers a big deal?
AI browsers act on the web on the user's behalf, querying, comparing and making decisions across multiple sites. The traffic that used to land on brand .coms is increasingly mediated by these agents. Brands that have only marketing copy for agents to read get summarised generically. Brands with structured product data and an agentic surface get answered specifically.
What role does the Model Context Protocol play?
MCP is one open standard that lets LLMs query brand data as a structured source. It is the agentic-web equivalent of an API contract: the brand defines what its data is and how to access it; the LLM uses that to ground its answers in primary information. We have written about MCP in more depth at /latest-thinking/model-context-protocol-mcp/.
Where do brands typically start?
Most start with one of three entry points. A conversational layer over existing product data on the brand's own site. A Digital Product Passport pilot on a single product line. Or an MCP integration that lets external agents query the brand's data with citations. The right entry depends on which conversation the brand can have credibly today and which competitive pressure is closest.
How we help
Where the agentic web lives in our work
Three engagement shapes for building a coherent surface across the layers.
TalkPod
When the product data exists but customers can't reach it conversationally. TalkPod connects to existing product data, configures for your range, and runs as a working pilot in one channel.
Explore TalkPodDigital Product & AI
When the question is what to build, where AI fits, and how the agentic surface wires into the rest of the product. Shape the brief, the data layer and the design choices.
View serviceDigital Strategy & Delivery
When the broader programme needs senior leadership inside the team to land it. Embed for the duration of the work and stay accountable to the outcome.
View serviceRelated thinking
Insight
Are agentic stores the future?
Swap launched the first dedicated agentic storefront recently, an early outlier that raises many strategic questions for brands and how they are accessed online.
Insight
When Products Can Answer Questions
Digital Product Passports start rolling out in 2027, attaching structured, verified data to millions of physical products. At the same time, the way people search for and evaluate those products is shifting, moving away from browsing and filtering towards asking questions and getting direct answers.
Insight
ChatGPT Apps: The Next Interface Shift
ChatGPT Apps marks a shift in how we use software. What this evolution means for people, businesses and the future of digital design.