---
title: "Dan Birman"
author: Dan Birman
url: https://dbirman.com/
updated: 2026-11-09
summary: "Designs how people work with AI agents: what they're allowed to do, when they ask, and how they earn more room. Home of Agent Trust Patterns."
---

# Dan Birman

# I design how people work with AI agents: what they’re allowed to do, when they ask, and how they earn more room.

I’m Dan, a designer in San Francisco, where it’s now. I run design at [Caplight](https://www.caplight.com), a private markets platform. Before that I designed tipping at Uber, AI legal research at Ruli, a care platform at Vida Health, and the Backbone One controller. I prototype in code. I’ve learned what trust takes where the stakes are real: money at Uber and Caplight, health at Vida Health, and legal research at Ruli.

## Now

Designing
:   Everything at Caplight, as its only designer: product, brand, and marketing.

Building
:   Agent Trust Patterns, a working library of interface patterns for AI that acts on people’s behalf.

Thinking about
:   What an AI agent should be allowed to do without asking, and how it earns more room.

## Agent Trust Patterns

AI is starting to act for people: spending, booking, signing, trading. What usually stops someone from handing a task over is the interface, more than the model. These are six patterns I think are missing. Each starts from a problem researchers have measured, proposes an interface that doesn’t exist yet, credits the closest prior work, and comes with a demo you can use.

1. [01**Consent by example**Don’t make people read the rules they’re granting. Show them the edge cases and let them sort.](https://dbirman.com/consent)
2. [02**Interrupt on drift, not on danger**The agreed plan is the baseline. Interrupt when the agent leaves it, not every time it does something risky.](https://dbirman.com/drift)
3. [03**Questions with a shelf life**Ask early, say what you’ll assume, and stop asking once the answer can’t help.](https://dbirman.com/shelf)
4. [04**Whose idea was this?**Before an irreversible step, mark any part of the reasoning that came from someone other than you.](https://dbirman.com/whose)
5. [05**Show the track record**Before you delegate, see how often “done” was actually done for this kind of task.](https://dbirman.com/record)
6. [06**The other side of the agent**Design for the person on the receiving end of an agent’s email, booking, or purchase.](https://dbirman.com/other)

## Agent Trust Reviews

Once a month I score an agent product people can use today against the six patterns. Each review is an eval: the same six tasks, with planted traps and a fixed grading table, run on the real product.

- Oct 2026[**ChatGPT Work**Its judgment is better than its interface. It usually makes the right call and rarely shows you what it decided.](https://dbirman.com/review-chatgpt-work)

## Selected work

- Ruli[**Getting lawyers past the blank prompt**The full case study: trust in AI answers, for people paid to doubt them.](https://dbirman.com/ruli)
- Uber[**Tipping without guilt**Trust around money between strangers. Riders tipped drivers $600M in the first year.](https://dbirman.com/uber)
- Vida Health[**Building a design team**Trust in care. The app’s rating rose a full star in a year.](https://dbirman.com/vida)
- Backbone[**Hardware and software, designed together**A new kind of controller, from sketch to launch.](https://dbirman.com/backbone)

## Writing

- Oct 2026[**Progressive reduction, ten years later**In 2015 I argued interfaces should step back as people learn. Now the agent is the one learning.](https://dbirman.com/reduction)
- Sep 2026[**How I design with AI**What changed when most of my design work moved into code.](https://dbirman.com/writing)

## Elsewhere

[About, talks, and history](https://dbirman.com/about). Email [hi@dbirman.com](mailto:hi@dbirman.com)

I’m happy to talk about advising, speaking, and anything in the patterns.