Agent Experience Optimization (AXO): How to Design Digital Experiences for AI Agents
Agent experience optimization is quickly becoming the discipline that separates brands that get chosen from brands that quietly get skipped, often by customers who never lay eyes on a website at all.
Picture a Tuesday night. A busy operations director asks her AI assistant to find a project management platform for a 40-person team, under $15 per seat, with SOC 2 compliance and a free trial, and then to start the trial. The assistant visits a dozen vendor sites in seconds. One hides pricing behind a “Contact sales” form. Another lists its compliance details only in a gated PDF. A third throws up a CAPTCHA. A fourth advertises a price “from $9” that doubles once the features she needs are added.
The agent starts a trial with the fifth vendor. The other four will never know they lost that buyer. There is no bounce to analyze, no abandoned-cart email to send, and no sales call to review. There is only silence in the funnel.
At JOS, we spend our days decoding the why behind customer behavior, and that silence concerns us more than any dip in traffic. It signals a new participant in the buying journey, one that reads your markup instead of admiring your hero image and has no patience for ambiguity. This guide explains what agent experience optimization (AXO) is, why the data says it matters now, and how forward-thinking leaders can prepare without abandoning the humans who still make the final call.
What Is Agent Experience Optimization (AXO)?
Agent experience optimization (AXO) is the practice of designing websites, apps, content, and APIs so that AI agents can reliably, safely, and efficiently discover, understand, and complete tasks on them.
An AI agent is software that acts on a person’s behalf. It browses, compares, fills in forms, starts trials, books appointments, and increasingly pays. AXO treats that agent as a genuine user segment with its own needs, constraints, and failure modes, just as good UX treats a first-time visitor or a screen reader user.
The companies building early tooling describe the work in practical terms. Snowplow frames AXO around three jobs: separating agent traffic from human and malicious bot traffic, tracking whether agents actually find products and check out, and understanding how agents and people collaborate toward a single goal. Form analytics firm Zuko calls it conversion rate optimization for AI-powered users.
Here is the simplest way to hold the idea. Being mentioned by an AI is visibility. Being usable by an AI is revenue. Our AI search optimization work helps large language models recommend your brand, and agent experience optimization ensures an agent can act upon arrival. The two are strongest when they are planned together.
Where Did the Idea of “Agent Experience” Come From?
The term “agent experience,” or AX, was coined by Netlify co-founder and CEO Mathias Biilmann in a January 2025 essay titled “Introducing AX: Why Agent Experience Matters.” He placed it in a lineage worth knowing. Don Norman popularized “user experience” in 1993, Jeremiah Lee introduced “developer experience” in 2011, and Biilmann argued that agents are the third class of user that software must now serve.
A year later, in One Year of AX, Biilmann organized the field into four areas: Access, Context, Tools, and Orchestration. The same post includes details most coverage skips. Netlify made AX its internal North Star and says it crossed 10 million developers as a result. Bessemer Venture Partners named AX the first of its AI developer laws. Sequoia’s Sonya Huang has described a shift from product-led growth to agent-led growth. The first job postings with “AX” in the title have also started to appear.
Biilmann is candid about the gaps. Outside developer tools, he writes, adoption across SaaS, e-commerce, and consumer products is still in its very early stages. His most quotable line doubles as a warning: “All products have an Agent Experience.” The only question is whether yours is good or bad.
Notice the pattern. Each new class of user, from people to developers to agents, rewarded the companies that studied it first and measured it honestly. That is why we believe agent experience optimization belongs in your marketing strategy, not only in your engineering backlog.
How Is Agent Experience Optimization Different From SEO, GEO and UX?
Agent experience optimization shares tools with its neighbors, but it answers a different question: can an agent actually finish the job here?
| Discipline | Designs for | Core question | Typical success metric |
| UX (user experience) | People | Can a human complete the task with ease? | Task success, satisfaction |
| DX (developer experience) | Developers | Can a developer integrate quickly? | Time to first successful API call |
| SEO (search engine optimization) | Search crawlers | Will we rank? | Organic traffic and qualified leads |
| GEO/AEO (generative and answer engine optimization) | AI answer engines | Will AI mention or cite us? | Share of AI mentions and citations |
| AXO (agent experience optimization) | AI agents acting for people | Can an agent complete the task here? | Agent task completion, agent-driven conversions |
Think of it as a relay. SEO and GEO get your brand onto the shortlist that a person or an AI assistant builds. AXO determines whether the agent can carry the baton over the finish line once it lands on your site, API, or checkout.
We made a related argument in SEO Without a Story Is Just Words on a Page: rankings alone have never closed a deal. Snowplow likewise cautions against treating AXO as merely the next generation of SEO, and the reasoning holds. Brands are not only being visited by third-party agents; many are building their own agents to replace parts of today’s human-facing experience. AXO sits where both meet.
Why Does Agent Experience Optimization Matter Right Now?
Agent experience optimization matters now because the web’s traffic mix has already shifted toward machines, and human clicks are shrinking in the places that once fed most websites.
Start with the headline number. Imperva’s 2025 Bad Bot Report found that automated traffic made up 51% of all web traffic in 2024, the first time in its records that bots outnumbered people. Malicious bots accounted for 37% and benign bots for 14%. A fair caveat: the figure comes from a single security vendor with a commercial interest in bot mitigation, and it counts every kind of automation. The direction, however, is echoed by Cloudflare, Akamai, and others.
Next, look at what happens to human clicks when AI sits between a person and the web. A Pew Research Center analysis of 68,879 Google searches by 900 U.S. adults found that people clicked a traditional result in 8% of visits when an AI summary appeared, compared with 15% without one. Links inside the AI summary itself were clicked in just 1% of visits.
Finally, the people who do arrive through AI are arriving in force and in a better frame of mind to buy. Adobe reported that traffic to U.S. retail sites from generative AI tools rose 693.4% year over year during the 2025 holiday season. Earlier, Adobe found that in October 2025, AI-referred visits generated 8% more revenue per session and bounced 31% less than non-AI traffic.
Put those findings together, and a clear picture emerges. The research phase of the customer journey is being handed to software, and people show up later, better informed and closer to a decision, if they show up at all. If your dashboard still celebrates raw sessions, it is measuring the wrong thing. As we explored in Your Website Traffic Means Nothing If Nobody Remembers You, volume without meaning is a vanity metric. The businesses that agents find easy to work with will capture a disproportionate share of the remaining demand.
How Do AI Agents Actually “See” and Use a Website?
AI agents interact with your business in four main ways, ranging from fragile to robust:
1. Reading raw content. The agent fetches your HTML or text and extracts facts. If key information appears only after heavy JavaScript runs, it may never be seen.
2. Driving the browser through page structure. The agent reads the DOM and the accessibility tree, the same labeled map of buttons, links, and fields a screen reader uses, and then clicks and types.
3. Looking at screenshots. When structure fails, the agent falls back to vision, interpreting pixels and guessing where to click. It works, but it is slower, costlier, and more error-prone.
4. Calling tools directly. The agent skips the interface and calls an API or protocol-based tool you expose. This is the most reliable path by a wide margin.
Where do agents stumble? In the same places impatient humans do, only faster and without complaint. Icon-only buttons with no label. Menus that appear only on hover. Surprise pop-ups. CAPTCHAs. Prices that say “starting at.” Policies scattered across three tabs and a PDF. Error messages that say “something went wrong” and nothing more. Bloated pages also waste the agent’s limited context window and bury the one fact it came to find.
Here is the insight that should reshape your roadmap: accessibility debt is agent debt. The WebAIM Million 2026 report found detectable WCAG failures on 95.9% of the top 1 million home pages, with an average of 56.1 errors per page. Six error types account for about 96% of them, including missing form labels, empty links, and empty buttons. Those are precisely the signals an agent reading the accessibility tree depends on to understand what a control does. Fix them, and you serve people with disabilities and AI agents in a single sprint.
What Are the Core Principles of Agent Experience Optimization?
Strong agent experience rests on six principles, which map neatly onto Biilmann’s Access, Context, Tools, and Orchestration framework:
1. Discoverability. Agents can learn what you offer and which actions are possible, without guessing.
2. Legibility. Content is structured, unambiguous, and light on noise, so the facts that matter are easy to extract.
3. Actionability. Core tasks can be completed through stable, well-defined paths, whether a clean form or an API call.
4. Predictability. The same input produces the same result, with clear states and specific, recoverable errors.
5. Trust and control. Agents can prove who they are and act within the permissions a person granted, while humans confirm anything high-stakes.
6. Efficiency. Tasks take as few steps, requests, and tokens as possible.
None of these is exotic. What changes is the standard of proof. A human will squint at a confusing page and push through. An agent will simply move on to your competitor. That is why we evaluate agent journeys the way our Deep Insight methodology evaluates human buyers: by observing actual behavior and asking why it happened, rather than assuming a page works because it looks polished.
What Can Agentic Commerce Teach Us About Agent Experience Optimization?
The most instructive AXO case study so far is a stumble, and it teaches more than any success story could.
In September 2025, OpenAI launched Instant Checkout in ChatGPT, powered by the Agentic Commerce Protocol it co-developed with Stripe. Walmart moved quickly, making roughly 200,000 products available from November. The results were sobering. Walmart’s Daniel Danker said purchases completed inside ChatGPT converted at one-third the rate of shoppers who clicked through to Walmart’s own site, and he called the experience “unsatisfying.”
The reasons are the lesson. CNBC reported that product selection remained limited and item information wasn’t always up to date. Forrester’s Emily Pfeiffer told CNBC that onboarding was arduous and that only about 30 Shopify merchants were live. By late March 2026, OpenAI said it would let merchants use their own checkout experiences and focus on product discovery. Forrester noted that this effectively turns the answer engine into a referral source, much like search.
Meanwhile, Google launched the Universal Commerce Protocol in January 2026, an open standard spanning discovery, purchase, and post-purchase support, designed to work alongside MCP, Agent2Agent, and Google’s Agent Payments Protocol.
Three takeaways for marketing and business leaders:
• Data freshness beats clever interfaces. Agents struggled with where inventory, pricing, and product details were stale. Accurate, near-real-time feeds are the foundation of agent experience optimization.
• Trust signals still matter at the moment of payment. A chat window with no visible cart or return policy made shoppers hesitate. The reputation you build through public relations, reviews, and client testimonials is part of what an agent and the person behind it weigh before committing.
• Don’t bet the business on one protocol. The standards are still shifting month to month. Build clean internal APIs you can map to whichever protocol wins in your category.
How Do You Measure Agent Experience Optimization Success?
You measure agent experience optimization the way you should measure any user experience: by whether the user, in this case, an agent, completes the task. Skip the vanity metrics and track these instead:
• Agent share of traffic, split by identified agents, agentic browsers and verified crawlers.
• Agent task completion rate for your revenue flows: search, add to cart, checkout, booking, and sign-up.
• Agent drop-off points and error rates, so you know exactly which step breaks.
• Tool and API usage, including success rates and latency for any MCP server or agent-facing endpoint.
• Representation accuracy, meaning whether AI assistants describe your prices, policies, and products correctly.
• Agent-attributed conversions and revenue.
Testing is refreshingly low-tech to start. Once a month, hand your five most important journeys to several popular agents and watch where they fail. Then automate it as a scheduled “agent eval.” A small tooling market is already forming: Netlify points developers to an open-source project called AXIS for measuring agent experience, and Zuko has launched an agent score focused on forms and checkout.
Where Should Your Company Start With Agent Experience Optimization?
Start with one journey, not the whole website. Pick the single task that generates the most revenue, then run it through this checklist:
• Can an agent find your core offer, price, and policies without running JavaScript?
• Is key information marked up with structured data?
• Does every button, link, and field have an accessible name?
• Can the task be completed in a linear, predictable flow?
• Are error messages specific enough to act on?
• Do you offer an API or MCP server for your core actions?
• Do your bot policies distinguish verified agents from malicious traffic?
• Can agents act with scoped, delegated permissions instead of passwords?
• Do payments and account changes require human confirmation?
• Are your product, price, and inventory feeds accurate in near real time?
• Have you watched an AI agent attempt this journey in the last 90 days?
The right scope depends on where you are. A founder just getting started can build agent-friendly foundations from day one. An established business usually gains the most by fixing its top two or three revenue journeys. Enterprise and middle market organizations need a cross-functional plan that connects marketing, engineering, and security.
If you remember one thing, make it this. Your next best customer may arrive as a few lines of code, sent by a busy person who trusted it to choose well. That person is still the one you are serving. Agent experience optimization is the discipline of making sure that when your agent comes knocking, your door opens, your facts are straight, and the job gets done.
At JOS, we help ambitious leaders replace guesswork with evidence. Our Deep Insight methodology decodes why buyers, and now their agents, make the choices they do, and our organic approach turns that understanding into a durable market position rather than short-lived spikes. If you’re ready to see how your brand performs for AI agents, request a consultation or review our pricing options. For more research-backed thinking, explore the JOS blog.
FAQs
Is agent experience optimization the same as AI search optimization?
No. AI search optimization, often called GEO or AEO, focuses on getting your brand mentioned and cited in AI-generated answers. Agent experience optimization focuses on whether an AI agent can complete a task, such as a purchase, booking, or sign-up, once it reaches you. Most brands need both.
Do I need an MCP server to be agent-friendly?
Not on day one. Clean semantic HTML, structured data, and an accessible interface solve most agent failures. An MCP server or documented API becomes valuable once agents regularly attempt tasks on your site and you want the most reliable path.
Should I block AI agents from my website?
Increasingly, blocking everything automated means blocking your own customers. A better approach is to verify agents where possible, rate-limit abuse, and require human confirmation for high-stakes actions.
How long does agent experience optimization take?
A focused audit of one revenue journey can surface meaningful fixes within a few weeks. Broader work, such as APIs, identity, and data feeds, typically unfolds over several quarters alongside your existing roadmap.
