Your best-qualified buyer may never load your website. Agent Experience (AX) is the practice of designing brand surfaces, content, and buying flows for AI agents acting on behalf of human B2B buyers, and it has moved from developer-tooling curiosity to a marketing problem in under two years.
Forrester’s 2026 buyer survey puts generative AI use in the B2B purchase process at 94%. Cloudflare’s audit of the open web found only 3.9% of sites serve machine-readable content when an agent asks for it. The distance between how buyers now research and what your site can actually answer is the entire opportunity.
Where the term came from
Agent Experience was coined by Matt Biilmann, CEO of Netlify, in early 2025. On the canonical agentexperience.ax site, Biilmann defines AX as “the holistic experience an AI agent will have as the user of a product or platform,” drawing a deliberate parallel with Don Norman’s concept of User Experience. The idea began in developer tooling, where coding agents were interacting with platforms such as Netlify with inconsistent success, before expanding to AI agents operating across broader digital products.
The term AX has also been used in other contexts:
- Augmented experience: An earlier use of the abbreviation.
- Agent experience in contact centers: The experience of human customer-service agents.
- Agent Experience in AI: The experience of AI software agents using digital products and platforms.
James Cadwallader also introduced the term independently at SEO Week 2025 in a content-optimization context. These strands have since converged around the emerging AI-agent definition.
AX is not a rebrand of UX
AX starts from a different premise. The user is an LLM-based agent with pattern-matching, text parsing, and explicit goal execution, not a human with visual perception, emotional response, and imprecise goals.
| Dimension | UX | AX |
|---|---|---|
| Who the user is | Human with visual and emotional perception | LLM-based agent with pattern matching and goal execution |
| Primary signals | Visual hierarchy, micro-interactions, accessibility | Semantic HTML, schema, machine-readable summaries, API discoverability |
| Design surfaces | Screens, click paths, onboarding flows | Structured data, API docs, MCP servers, plain-text pricing |
| Success metric | Task completion, satisfaction, time-on-page | Recommendation inclusion, citation accuracy, agent task completion |
| Failure mode | Confusion, abandonment | Hallucination, silent omission, misclassification |
| Trust mechanism | Brand perception, visual credibility | Verifiable structured data, cross-source consistency |
Where AX inherits from UX: Clear information architecture, logical navigation, and unambiguous content help both audiences. AX teams describe the dependency bluntly: AX inherits every decision made at the UX level, whether or not that decision was deliberate. Where UX is vague, AX becomes unreliable.
Where AX diverges:
- UX can lean on visual persuasion, clever copy, and aesthetic engagement. Agents cannot be charmed. They parse semantic meaning.
- Content built to move humans, meaning vague superlatives, narrative-first pricing pages, and JavaScript-rendered specifications, fails them systematically.
- Biilmann’s most-quoted example is a single line of improved CLI output that took Claude Code’s success rate on Netlify from frequent failure to reliable deploys. Same product, different Agent Experience.
The evidence agents are doing buyer-side work
Generative AI use in B2B purchase research rose from 89% to 94% across Forrester’s 2025 and 2026 buyer surveys, with the latter covering nearly 18,000 global business buyers. More than half of buyers now compare vendors inside AI tools before contacting them, while close to half use AI to help build the internal business case. The 6sense buyer report, based on roughly 4,000 buyers across North America, EMEA, and APAC, also puts LLM use at 94%.
Memetik’s analysis of more than 50,000 B2B buying journeys shows how that behaviour is changing the early buying process:
- Buyers run an average of 4.7 AI queries before their first vendor website visit, up from 0.3 in 2023.
- 63% use an AI assistant to compare vendors before building a shortlist.
- The average consideration set has narrowed from 12 vendors to three or four.
- 58% of software purchases under $50,000 are completed without the buyer visiting a vendor website.
Agent activity is also extending beyond research. Production use cases now include RFP analysis, quote management with live inventory checks, pricing discovery, and first-pass contract review. Vendor roadmaps reflect that expansion, with Anthropic, AWS, and Google building agent-facing capabilities into enterprise products. OpenAI’s in-chat purchasing experiment also exposed an important boundary: AI can increasingly handle discovery and evaluation, while payment and transaction responsibility remain harder to delegate.
The standards layer that formed in 2026
The standards layer that formed in 2026
Agent-to-merchant interaction is starting to move from vendor-specific implementations toward shared protocols. Three major efforts emerged within a few months of one another, though their relevance to B2B varies.
| Protocol | What it covers | B2B relevance |
| Agentic Commerce Protocol | Discovery inside an AI assistant, with checkout returned to the merchant | Primarily retail-focused |
| Universal Commerce Protocol | Discovery, capability negotiation, checkout, and post-purchase handoff | Primarily retail-focused |
| Agent Payments Protocol (AP2) | Signed mandates covering user intent, agent actions, and the final charge | Stronger enterprise relevance because it maps more closely to approval and procurement workflows |
OpenAI developed the Agentic Commerce Protocol with Stripe, while Google and Shopify introduced the Universal Commerce Protocol at NRF in January 2026. UCP is already used across Google AI Mode, Gemini, and YouTube Shopping. Both point toward standardized agent commerce, but their current design is primarily consumer and retail-oriented.
Google’s Agent Payments Protocol has clearer enterprise carry-over. Each transaction is represented through three signed mandates covering what the user requested, what the agent assembled, and what will ultimately be charged. That structure resembles enterprise approval chains more closely than a conventional checkout flow. Google released a second revision in April 2026 and transferred the protocol to the FIDO Alliance.
For marketing teams, the implication is simpler: agent-facing surfaces are becoming standardized and increasingly measurable. Agent readiness can therefore become something teams evaluate against defined technical requirements rather than treating it as an abstract concept.
What this does to the funnel
Traditional B2B attribution captures less than 30% of the buying journey. Bain describes the growing role of AI as marketing’s new middleman: discovery, evaluation, and shortlisting increasingly happen inside AI tools before a buyer reaches a vendor site. Its data also shows generative AI traffic to vendor websites growing sharply between mid-2024 and early 2025 even as overall site traffic declined.
That changes the role of sales as well. The same 6sense research finds that four out of five deals go to the vendor the buyer already preferred before first contact. More of the competitive decision therefore happens upstream of the form fill, inside the ghost funnel that conventional click-based attribution cannot fully capture.
Three metrics help measure what happens there:
- Citation share: How often AI systems cite or recommend your brand for relevant category prompts.
- Brand entity authority: How consistently models recognize your brand as an entity with accurate attributes.
- Agent-completion rate: How often an agent can successfully retrieve or interpret information such as pricing, features, and integration documentation.
ReSO’s Visibility Audit tracks the first two alongside the broader B2B measurement surface.
The case for caution
Several factors still limit how far Agent Experience can extend into fully autonomous B2B buying:
- Autonomous procurement remains limited: Full AI integration into enterprise source-to-pay systems is still in the single digits. Trust, negotiation, contractual obligations, and regulatory requirements continue to require human involvement.
- Hallucination remains a material risk: AI systems can fabricate citations, contract terms, pricing information, or product capabilities, creating obvious problems in procurement workflows.
- Attribution is still imperfect: Some of the “invisible” buying journey attributed to AI may simply be anonymous human research, which existed long before generative AI. AX can address agent-driven discovery, but not all unattributed activity.
- Governance still requires people: Even when an agent identifies or shortlists a vendor, enterprise purchases still move through approval, legal, security, and multi-stakeholder review.
These constraints affect the pace of adoption more than the direction of travel. AI-mediated buying is becoming more common; the bigger uncertainty is when fully agentic procurement moves from isolated deployments into standard enterprise practice.
Access, context, tools, and orchestration
Biilmann breaks Agent Experience into four practical areas:
- Access: Can an agent reach the product through the right APIs, OAuth scopes, keys, and public endpoints?
- Context: Does it have enough documentation, structured information, and semantic markup to understand what the product does and how to use it?
- Tools: Are APIs, CLIs, and SDKs designed for agent workflows, with predictable commands, machine-readable output, and clear error messages?
- Orchestration: Can agent actions be triggered, coordinated, and completed inside the product rather than handled as isolated requests?
Cloudflare’s Agent Readiness score turns a similar idea into a measurable benchmark. Launched in April 2026, it scores sites from 0 to 100 across discoverability, content, bot access control, and capabilities. It also checks agentic commerce standards, although those do not currently contribute to the score.
The benchmark is still low. Cloudflare found that only 3.9% of sites serve markdown when an agent requests it, while just 4% declare AI usage preferences in robots.txt. For teams working on AX, that creates a relatively clear starting point: improve how easily agents can access, interpret, and act on the information already available across the site.
Want to see where your site stands? Run an agent-readiness and AI visibility audit to identify the surfaces that may be limiting how AI systems discover, interpret, and recommend your brand.
Frequently Asked Questions
Is AX a rebrand of generative engine optimization?
No, GEO and AEO focus mainly on earning citations in AI-generated answers through content and entity optimization. AX is broader, covering schema, APIs, CLI output, pricing-page structure, and other surfaces agents consume. It also considers whether an agent can complete a task, not just find you in AI search results.
Does AX replace UX?
No, AX builds on UX. Clear information architecture, structured data, and unambiguous content help both humans and agents. Poor UX decisions can also weaken AX because agents depend on the same underlying structure.
Where should a team start?
Start with a complete schema, machine-readable pricing and feature pages, and markdown content negotiation. Then review API documentation, CLI output, and error messages for agent readability. An agent-readiness scan can help identify the highest-priority gaps.
Is the agentic-buyer thesis playing out, or is it overhyped?
AI-assisted buying is already growing, but full autonomy remains limited. Most purchases still involve human decision-making. The marketing implications are immediate, while fully agentic procurement is more likely to develop over the next two to three years.
