Ravi Mehta Added Agents to the Definition of UX. The Second User Never Sees the Screen.

Ravi Mehta's UX definition now says people and agents. LukeW, Resend, and Lemkin showed what that means. Read the agent's requests, not the screen.

Falk Gottlob · · 10 min read

AI Agents Falkster cover on green: a cream browser window with a navy slider control inside it, and beside it a navy robot turned away from the window, reading a cream sheet of paper with three navy bars like a table, with an orange arrow curving from the robot into the sheet.

Ravi Mehta published the sixth competency in his Top 1% Builder series on November 12, user experience design, and the definition has a word in it that would not have been there a year ago. This is a Rewrite, and what gets rewritten is not the definition. It is the method underneath it.

The short version

  • Ravi Mehta's November 12, 2026 definition of user experience design is the ability to apply usability principles, design patterns, and interaction models so that people and agents can access a product's full value.
  • The same week, three other posts on my roster described the agent as a user: Luke Wroblewski shipped his 17,000-question Ask LukeW archive as an MCP server so the reader never visits, Zeno Rocha of Resend described a pricing page that serves a slider to people and a table to the agent that adds .md to the URL, and Jason Lemkin cancelled Canva after 12 years because his agents moved the work to Gamma and Higgsfield and never looked back.
  • Mehta's method, watch users work and look for four kinds of friction, finds the first user and not the second, because the agent never sees the screen.
  • The rewrite is to read the agent's requests, with one test: can an agent that has never seen the product get the fact it came for in one request, with no screen in between.
  • My own scan is that test run from the reader's side: 165 sources, 72 answer a script in one request and gave me 229 posts this week, 93 publish where a script cannot read them and gave me nothing.

What Mehta wrote

The post opens with a Tripadvisor filter redesign, visually striking and icon-heavy, that tested at "a 20% drop in conversion, a potential $100M revenue hit if we rolled it out." Customers only benefit from value they can reach. From there the definition: user experience design is "the ability to apply usability principles, design patterns, and interaction models," on your own or with design partners, to make it easy for people and agents to access a product's full value.

He means the word. Granola-style notetakers, he writes, "worked by having an agent join the video call like any other attendee." Later: "The same care applies when an agent uses your product." And: "The experience is the whole path, from a person or an agent to your product's value."

His method for finding the work is to watch users and look for four kinds of friction: "Physical effort," "Cognitive burden," "Translation work," and "Workarounds." Then "Count the work after the AI responds, too." His ladder runs from the APM who learns "to assess an experience and recommend concrete improvements" to the VP or CPO who fosters "a design-driven culture." He closes: "Before adding another room to the house, check whether the front door is locked."

I have no quarrel with the definition. The method is built for one of its two users.

Three posts that said the same thing without the word

Luke Wroblewski, November 11. Ask LukeW has answered more than 17,000 questions about digital product design on his site. Each person had to come to the site to ask. Now the same advice is an MCP server at lukew.com/mcp, with keyword and semantic search exposed as tools so the model decides which to use, every result linking back to its source. His line about the readers: going forward, they "won't have to" visit. He also built an Interface Reviewer agent on top of it that reviews a signup form and writes the prompts a coding agent can apply. Decades of interface writing, and the form he calls "perhaps the most useful" is the one where no human reads the page.

Zeno Rocha, co-founder and CEO of Resend, on Latent Space on November 10. Resend's pricing page has a slider for humans. "Add .md to the URL and the agent gets a table it can read in one request." He cites an Amplifying.ai study in which Claude Code recommended Resend first in 64 of 102 extractable recommendations, with SendGrid first 7 times and named as an alternative 55 times, and he is candid that Latent Space's own broader study put Postmark top for most models. Resend's MCP server grew from roughly 20 tools to more than 80, because he wants a chat-client user to be able to do everything the dashboard allows. The company had 4.3 million users and 120,000 paying customers at recording.

Jason Lemkin, November 11. He and his colleague Amelia cancelled Canva after about 12 years. Nobody was angry and nobody decided to replace it. Their agents moved image and video work to Higgsfield and decks to Gamma, and never reconsidered. His first learning from the week: "Agents pick vendors once and then stop looking."

Four people, one week, four angles on one fact. The product now has a second user, and the second user never sees the screen.

Why the method misses the second user

Mehta's four frictions are the right list for a person. Physical effort is a click too many. Cognitive burden is a label that makes you think. Translation work is turning your goal into the software's commands. Workarounds are the spreadsheet next to the app.

An agent shows none of those. It does not click, it does not read labels for meaning, and it does not keep a spreadsheet. It issues a request to a URL and reads what comes back. If the fact it came for is in the response, it uses it. If the response is a page built for a human eye, a slider, a login wall, a chart with the number drawn in, it either calls something else or it moves on. Rocha's .md suffix is the fix for that. Lemkin's cancelled subscription is what the absence of the fix looks like from the buyer's side, and the buyer did not even know a decision was being made.

So you cannot watch the second user work. There is nothing to watch. You can read its requests, which is a different discipline, and the design handbook has a chapter on it, Reading Behavior You Did Not Design(coming Dec 23). The request log is the usability study for the agent, and most teams have never opened it.

The rewrite

Keep Mehta's definition. Replace the method for the second user with one test.

Can an agent that has never seen your product get the fact it came for in one request, with no screen in between?

Run it on the pages an agent needs first: the price, the limit, the status, the record behind the thing a person sees on screen. For each, make one request as the agent would and read what comes back. If the answer is in the response, that page passes. If the answer is in a slider, a drawn chart, or behind a wall, it fails, and the fix is usually the one Rocha describes, the same fact served as text at the same URL.

It belongs on Mehta's ladder at the first rung, next to "assess an experience and recommend concrete improvements," because it is the cheapest assessment on his list. No lab, no recruit, no session. One request and a read. This is the agent-interface half of What Design Owns Now: design specifies behavior, and for the second user the behavior is what the URL returns.

The same test from the reader's side

I run this test every morning without meaning to. My weekly creator scan reads 165 sources: newsletters, company blogs, and the people I follow for product, design, AI, and forward deployed engineering.

This week, 72 of them answered a script in one request each. Those 72 produced 229 posts in the window, and 95 of them scored as on-thesis and went into the digest that this post came out of. The other 93 publish only where a script cannot read them, 92 of them on LinkedIn and one on a Kit signup page. They produced nothing, except one edition I opened by hand in a browser because the person is worth the trip.

Nobody at those 93 decided to be unreadable to me. They published where their audience was, and my reader is a script that comes once. That is exactly what happened to Canva. Nobody at Canva decided to lose a 12-year customer. The customer's reader changed, and the reader never opened the app.

I wrote about the other end of this in Reddit Fell 29% on 'AI Mode Does Not Drive Traffic.' Notion's Top Channel Is an AI Tool.: the referrer moved from the results page to the assistant, and the assistant sends people to the page it can read. falkster.com made its choice in September. Markdown at every URL, llms.txt, agents.md, and a downloadable corpus, so the second user gets the substance in one request. The Enterprise AI Agents thesis is that agent count is a vanity metric and what matters is how many agents still complete production work after ninety days. An agent that cannot read your product does not complete work with it, and nobody will tell you. The one-request test is how you find out before the renewal does.

Mehta closes with the front door. I argued last month in Personal Agents Will Call Every Few Hours. Kill the Front Door. that the front door is the wrong place to meet an agent at all. For the second user there is no door. There is a URL, and a request that comes once.

This week

Pick the three facts an agent would need from your product before it recommends you: the price, the limit, the current status. For each, make one request to the URL a person would use, as a script, and read what comes back. Count how many of the three are in the response as text. That number is your second user's conversion rate, and it is cheaper to find than Tripadvisor's 20%.

Related answer: How do you design a product for an AI agent as the user?

Sources: Ravi Mehta, "User Experience Design.," Top 1% Builder with Ravi, November 12, 2026, read in full; the Tripadvisor 20% and $100M are his. Luke Wroblewski, "Ask LukeW: in Your AI Tools," November 11, 2026, read in full; the 17,000 is his. Zeno Rocha with swyx, "Designing for Agents," Latent Space, November 10, 2026, article and show notes read in full, the episode not listened to; the 64 of 102, 7, 55, 20 to 80 tools, 4.3 million, and 120,000 are as the article reports them, and the Amplifying.ai and Latent Space studies were not opened. Jason Lemkin, "Our Agents Never Picked Canva," SaaStr, November 11, 2026, read in full; the 12 years is his. The 165, 72, 93, 92, 229, and 95 are my own scan's counts for the week of November 5 to 12, 2026.

72 of 165

sources on my roster that answer a script in one request. The other 93 gave this week's scan nothing.
Share this post

Frequently asked

What is Ravi Mehta's definition of user experience design?+

In the sixth post of his Product Competency Toolkit series, published November 12, 2026, Ravi Mehta defines User Experience Design as the ability to apply usability principles, design patterns, and interaction models, independently or with design partners, so that people and agents can access a product's full value. The word agents is the change, and he repeats it: the same care applies when an agent uses your product, and the experience is the whole path from a person or an agent to the value.

What do LukeW, Resend, and Lemkin have to do with it?+

They described the agent as a user the same week. Luke Wroblewski shipped his 17,000-question Ask LukeW archive as an MCP server so the reader never visits his site. Zeno Rocha of Resend described a pricing page with a slider for people and a .md version the agent reads as a table in one request, and cited a study in which Claude Code recommended Resend first in 64 of 102 cases. Jason Lemkin cancelled Canva after 12 years because his agents moved the work to Higgsfield and Gamma and never reconsidered.

Why rewrite Mehta's method instead of his definition?+

The definition is right. The method under it, watch users work and look for physical effort, cognitive burden, translation work, and workarounds, only finds the first user. An agent never sees the slider, the icon, or the front door. It reads whatever the URL returns and either gets the fact in one request or moves on, which is what Rocha's .md suffix and Lemkin's cancelled subscription both describe.

What is the one-request test?+

Can an agent that has never seen your product get the fact it came for in one request, with no screen in between? Run it on the pages an agent would need first: pricing, limits, status, the record behind the thing a person sees. It belongs at the first rung of Mehta's ladder, next to assessing an experience, because it is the cheapest assessment on the list.

What is the 72 of 165 number?+

My weekly creator scan reads 165 sources. This week 72 answered a script in one request each and produced 229 posts, 95 of them on-thesis. The other 93 publish only where a script cannot read them, 92 on LinkedIn and one on a Kit signup page, and produced nothing for the digest except one edition I read by hand in a browser. That is the agent's side of the market Lemkin described from the buyer's side.

What has falkster.com done about its own second user?+

Since September the site serves markdown at every URL, publishes llms.txt and agents.md, and packages the practice as a downloadable corpus, so an assistant can get the substance in one request. The one-request test is the rubric row that keeps that true as the site grows.

Ravi MehtaLuke WroblewskiZeno RochaResendJason LemkinCanvaGammaMCP
Helpful?

Keep reading

The short answer

PM AI Agent Fleet, Mapped to the 7-Stage Operating SystemA live index of every AI agent for product managers, mapped to the 7 stages of the PM Operating System: Sense, Discover, Decide, Build, Ship, Measure, Amplify.Read the chapter

Part of

Enterprise AI Agents

About the author

Falk Gottlob

Falk Gottlob

Product Executive · Founder, Falkster.AI

Thirty years shipping product, from Microsoft Research and Adobe to Salesforce, where he grew Quip into what became Slack Canvas. Four startup exits, including a $6.5B healthcare merger and a company Microsoft bought. Four-time Chief Product Officer. Now founder of Falkster.AI, building Heidi, the company brain for AI-native businesses. This notebook is written from inside the build, not above it.

Sign in with LinkedIn to leave a comment.

Sign in with LinkedIn