How are you making your product more agent-friendly?

[zenorocha]

I've been thinking a lot about how to ensure our SaaS product can be used by agents instead of humans.

Even things like reCAPTCHA, which we added to prevent bots from signing up, are now being viewed differently.

One interesting thing I've seen today is people comparing how competing products can perform the same task as an agent.

• Resend vs Loops: https://x.com/dylibso/status/1899940324413411338
• Supabase vs Neon: https://x.com/dylibso/status/1897349465717293544

How are you testing if your product is agent-friendly?

[sean-roberts]

@zenorocha This is a great question and something I'm looking at a lot.

To help ground this work, folks in the AX group have established some initial principles of AX These help identify what's important and what we should be striving for.

From there, it's been straightforward to manually assess where things are.

• what features matter for AX?
• assess those features...
  • do those features have APIs?
  • do those APIs have context?
  • manually validating with relevant tools the quality of that context
  • are they usable without human intervention? If not, is this absolutely required?
    • the captcha question is important here.. captcha is probably important but where it shows up in the flows might be different for agent flows to reduce the amount of human intervention needed
  • do APIs have feedback to understand the success or status of API usage?

I'm thinking about potentially codifying these into a checklist or some form of "manual assessment" (would that be useful to you?).

This research is also driving research into automating all of this. So starting with unscalable/manual mechanisms as we deliver open ways to automate assessment.

[nobilix]

We’re developing an llmtxt plugin for PayloadCMS and are first integrating it into our website.

I see at least three ways to make a website or service more AX-friendly:
1. Enhancing machine readability, including sitemaps and llmtxt integration.
2. Providing a public API for better accessibility and interaction.
3. Implementing an MCP server to improve structured data access.

[Marcwab]

After testing a client website (Hotel chain), I found that cookie consent banners and popups are a significant barrier to agent navigation. In many cases the cookie banners and other popups (Login, Offers, "X people are viewing", etc...) couldn't be dismissed reliably through standard selectors, creating some kind of agent trap.
Here are my proposed agent-friendly solutions:

• Create endpoints (e.g., /api/consent/agent) for programmatic consent management
• Allow consent preferences to be set via request headers without UI interaction
• Implement consistent, stable selectors with machine-readable attributes
• Detect agent interactions via headers and provide simplified flows (without popups)