Back
Product
What Is Agent Experience (AX), and Why Will It Decide Who Wins?
Agent experience (AX) is how easily an AI agent can discover, understand, and use your product. Learn why it matters and how to measure it.
Scope
From the Office
The rise of agent experience (AX)
Most companies still treat AX as an engineering curiosity, and the ground is moving faster than that.
People used to find, evaluate, and install software themselves. Now agents like Claude Code, Codex, and Cursor do all three on their behalf, often before a human ever looks at what’s happening. That shift is already showing up in the traffic hitting your docs, your API, and your onboarding flow. How well your product works for these agents has a name: agent experience, or AX. It’s how easily an agent can discover, understand, and use your product well enough to complete a task.
This isn’t a prediction. G2’s 2026 Buyer Behavior Report found that 82% of B2B software buyers had sourced product recommendations from an AI chatbot within the prior two years. 61% were already using or planning to use AI agents somewhere in their buying process, and another 19% were open to it for specific tasks.
UX was built for human users and DX for developers. AX is built for agents, and it’s becoming the gatekeeper between you and an entirely new category of customer. That customer doesn’t wait around. When an agent gets stuck, it quietly leaves and picks a competitor.
The churn you can’t see
In G2’s Answer Economy survey, 69% of buyers said an AI chatbot led them to pick a different vendor than they had planned, and 33% bought from one they had never heard of. If a recommendation can redirect a first purchase that easily, it can redirect a renewal too.
The unsettling part is the silence. A customer who leaves usually makes noise first: a complaint, a support ticket, a cancellation call. An agent makes none. It hits a wall in your setup, tries a competitor, and never comes back. Your dashboards show nothing wrong the whole time. By the time the loss shows up in revenue, the decision was made months ago, by something you never knew was evaluating you.
The two sides of agentic experience: discoverability and usability
Picture a developer telling their agent, “Add payments to my app.” Whether you win that customer comes down to two questions.
The first is Discoverability: can the agent find you, and does it choose you? The agent weighs the options it knows and picks one, often before the developer sees a shortlist. If you sell payments, you should know how often agents pick you for a prompt like that, and why they pick someone else when they don’t.
The second is Usability: once the agent has chosen you, can it finish the job alone? It has to create an account, get API keys, run a test charge, and recover when something errors. If your signup requires email verification, or your error message only says “request failed,” the agent stalls and the developer ends up with a competitor’s integration.
Together these cover the full lifecycle: discover, evaluate, set up, execute, and recover from errors.
A single weak link anywhere in this process can lose you the customer for good. This isn’t a new feature that is simply nice to have, it’s a pass/fail gate directly affecting your revenue.
What trips agents up and how to fix it
Most AX failures come from a handful of fixable things.
Discoverability: why agents don’t pick you
Your docs need a browser. If your docs or pricing render with JavaScript or sit behind a login, an agent may see an empty page and move on to a competitor it can read. Fix: serve them as plain text or markdown.
You only exist on your own site. Agents draw on review sites, comparisons, GitHub, and forums, not only your homepage. If you’re thin or out of date in those places, you don’t make the shortlist. Fix: keep your presence current wherever agents read about your category.
You don’t say what you do. Agents match a request like “add payments to my app” against what your product says it does. A tagline like “the all-in-one platform for growth” gives them nothing to match. Fix: say plainly which tasks your product handles, and link a quickstart for each.
Usability: why agents don’t finish
Failures that look like success. A setup step fails but reports that it worked, so the agent carries on and breaks later, far from the real cause. Fix: make every step report failure clearly, through exit codes and HTTP status codes, and give the agent a way to verify setup.
Steps that wait for a human. A confirmation prompt or browser login leaves the agent stuck until it times out, and nothing gets logged. Fix: offer a non-interactive path for everything, including auth.
Retries that make things worse. Agents retry when something times out. If a repeated request creates a duplicate or trips your rate limit, the agent digs itself deeper. Fix: make requests safe to repeat, and tell the agent when to try again.
Good AX is a clean path from “what should I use?” to “task complete.” Bad AX can still get you chosen, but you'll still lose the customer at setup or the first error.
How to measure agent experience
Your dashboards show nothing wrong because they only track people. To see what agents do, run them through real tasks and start by tracking these three numbers:
Selection rate: how often agents pick you when given options.
Completion rate: how often they finish a task once they start.
Drop-off point: where they stall when they don’t.
Selection rate measures discoverability, and the other two measure usability. Run the same tasks across Claude Code, Codex, and Cursor, because they fail in different places and a fix for one may do nothing for another.
The clock is already ticking
Most companies still treat AX as an engineering curiosity, and the ground is moving faster than that. In June 2026, Cloudflare reported that bots had passed humans in requests for web pages across its network, 57.5% to 42.5%, well ahead of its CEO’s own forecast. MCP, the standard agents use to connect to tools, went from launch to more than 10,000 public servers in about a year.
As more decisions get delegated to agents, AX will matter to growth and product teams as much as to engineers. The companies investing now get easier for agents to find and use with every release, and that head start gets harder to close.