Ax-check: Test Whether Your Product Can Be Used by AI Agents

When AI Agents Become Your New Users

For the past decade, the central question in product design has been "how do humans use software?" Now, as AI agents rise rapidly, a new question has emerged: Can your product actually be used by an AI agent?

A recent Show HN project on Hacker News — Ax-check.com (Agent Experience Check) — is aimed squarely at this emerging need. It tries to answer an increasingly practical question: when automated programs and LLM-powered agents attempt to access, understand, and operate your website or product, can they successfully complete their tasks?

The project received 12 upvotes and 11 comments on HN — not a viral hit, but it touches on an issue that sits right at the critical inflection point of real-world AI deployment.

What Is "Agent Experience"?

If UX (User Experience) is about how smoothly human users interact with a product, then AX (Agent Experience) is about the usability of a product when automated agents interact with it.

The evaluation criteria are fundamentally different. Human users are good at handling ambiguous information, recognizing visual cues, and tolerating a certain degree of interface chaos. AI agents, on the other hand, rely heavily on structured data, clear semantic markup, stable and predictable interfaces, and machine-readable documentation.

Why This Is Starting to Matter

As AI agents are increasingly tasked with "completing actions on behalf of humans" — such as automatic booking, price comparison, form filling, and API calls — whether a product is agent-friendly directly determines whether it can be integrated into these automated workflows.

In other words, a brand new source of "traffic" and "conversion" may emerge: not a person clicking through, but an agent completing the action for them. If your product isn't agent-friendly, you'll be shut out of this new lane entirely.

In terms of technical implementation, AI agents generally fall into two categories. The first is browser automation-based agents (using tools like Playwright, Puppeteer, or Computer Use), which "see" pages and click elements just like a human would. The second is API/structured interface-based agents, which call server-side interfaces directly to complete tasks. The former is highly sensitive to DOM structure, semantic tags, and visual layout — CAPTCHAs, dynamic rendering, and poorly labeled buttons can all cause task failures. The latter requires clear API documentation, authentication mechanisms that are friendly to programmatic calls, and error responses in a parseable format. The two types of agents have different requirements for "product usability," which is why "agent-friendliness" is hard to measure by a single standard — product teams first need to identify which type of agent they're targeting before optimizing accordingly.

The Pain Points Ax-check Aims to Solve

From the project's positioning, Ax-check.com offers a diagnostic capability: helping product owners check how their website or service performs when faced with AI agents.

Typical inspection dimensions might include:

• Whether page content is structured and semantically clear
• Whether key action paths can be programmatically identified and executed
• Whether there are blockers like excessive JavaScript rendering or CAPTCHAs that hinder automation
• Whether documentation and interfaces are machine-readable
• Whether robots rules and access policies are clearly defined for agents

For developers and product teams, the value of this type of tool lies in transforming the vague concept of "agent usability" into concrete, measurable, and optimizable metrics.

Points of Disagreement in Community Discussion

Show HN projects like this typically spark discussion in a few directions, and Ax-check is no exception.

One camp argues that "agent-friendliness" is becoming the next essential capability after "mobile-friendliness" and "SEO-friendliness" — products that prepare early will gain a first-mover advantage. Just as responsive design went from a nice-to-have to a baseline requirement, AX could follow the same path.

The other side is more cautious: behavioral standards for AI agents are not yet unified, and different models and frameworks interact very differently. Optimizing for one type of agent won't necessarily work for all of them. There's also concern that over-optimizing for agents could introduce new security and abuse risks — after all, making it easier for machines to operate your product also lowers the bar for automated scraping and bulk operations.

The Paradigm Shift from SEO to AXO

Zooming out, what Ax-check represents is a larger trend: the "audience" for content and products is expanding from purely human to a hybrid structure of humans + AI agents.

In the past, we did SEO (Search Engine Optimization) to help search engines better understand and index content. Today, a similar concept — call it AXO (Agent Experience Optimization) — is emerging. It demands that products not only look good and feel intuitive, but also be easily understood and callable by machines.

This means structured data, clear APIs, semantic markup, and stable interface contracts will evolve from "technical best practices" into real, tangible competitive advantages.

Structured data plays a foundational role in this trend. Semantic markup defined by Schema.org (such as JSON-LD types like Product, Event, and FAQPage) was originally designed for search engines, but it equally helps AI agents quickly understand the meaning and relationships of elements on a page — without relying on natural language inference. Similarly, the OpenAPI Specification (formerly Swagger) provides machine-readable standard description files for REST APIs, allowing agents to automatically discover interface capabilities, parameter formats, and authentication methods. Anthropic's Model Context Protocol (MCP) is a newer attempt — it defines a standard interaction protocol between AI models and external tools/data sources, with the goal of enabling agents from different vendors to invoke product capabilities in a unified way. The maturity of these technical standards will directly determine whether "AXO" can develop actionable best practices the way SEO has.

Conclusion: An Early Signal Worth Watching

As an early-stage project, Ax-check.com currently has limited community traction, and its diagnostic methodology, coverage of agent types, and result accuracy all still need to be validated through broader real-world use. But the question it raises is genuinely forward-looking.

As more and more tasks are delegated to AI agents, "can your product be used by an agent" will no longer be a technical footnote — it could become the critical threshold that determines whether a product can enter the next wave of automation. For product and engineering teams, starting to think about agent experience now might not be too early at all.