# Most of Your Docs Readers Are Now AI Agents

Hahnbee Lee, Mintlify & Lincoln Murr, Coinbase | AI Engineer NYC 2026 | 18:52

Source: https://www.youtube.com/watch?v=YX4Z4XqZ9Vo
Channel: AI Engineer (https://www.youtube.com/@aiDotEngineer). Summarised by AIE Talks.
Page: https://aietalks.com/talks/most-of-your-docs-readers-are-now-ai-agents
Published: 2026-11-01
Tags: agents, developer-experience, finance

## TL;DR
- Documentation for agents must prioritize small page sizes, accurate content, and current information over visual design.
- Coinbase is exposing its business through APIs so agents can trade, manage money, and pay for other services with appropriate limits.
- Giving agents access to money can let them purchase APIs and services independently, which increases their ability to run for longer without human intervention.

## Summary
Hahnbee Lee and Lincoln Murr discuss what changes when AI agents become major users of software. Lee says Mintlify has seen agent traffic rise from about 10% to about 80% of visits this year. For agents, page size, accuracy, and fresh content matter more than branding or visual design. Murr describes Coinbase's plan to make its functionality available through APIs, with controls for authentication, security, and spending. He argues that agents need money to act independently, including paying for APIs, paywalled information, and services such as lawyers or accountants. They also discuss crypto payments, including the x402 standard and its use for very small transactions. Lee expects technical documentation work to become more automated, while human judgment remains important for product insight and narrative writing. She closes by warning teams to consider maintenance before building internal tools for fast-moving agent systems.

## Key ideas
### Agents stop when documentation creates friction
[01:21](https://www.youtube.com/watch?v=YX4Z4XqZ9Vo&t=81s)
Hahnbee Lee says documentation used to put more weight on design and the front end. Agents do not care about a site's branding or visual presentation. They care about whether pages fit within their context window and whether the instructions are accurate. If an agent encounters a small friction point while setting up a product, it may stop, attempt brute force, or switch to another product. Lee also says teams need automation that keeps documentation from going stale because engineering is moving faster and the rest of the organization has to keep up.

### Coinbase is exposing its business through APIs
[03:35](https://www.youtube.com/watch?v=YX4Z4XqZ9Vo&t=215s)
Lincoln Murr says Coinbase is pursuing two related goals. It wants to build products specifically for agents, while also making its existing trading and money-management functions available to them. Browser use is possible, but it brings authentication and security problems, especially when an agent can access finances. Coinbase wants APIs with limits and guardrails. Murr compares this shift to the move from web software to mobile software, except that mobile access is now being extended to agents.

### Crypto payments make tiny, global agent purchases possible
[05:30](https://www.youtube.com/watch?v=YX4Z4XqZ9Vo&t=330s)
Murr says crypto is useful for agent payments because digital dollars can move cheaply, quickly, and without permission from an intermediary. He points to microtransactions as a particular use case. An agent could pay a fraction of a cent to access an API, a service, or a paywalled site. He describes x402 as an internet-native payment standard that has been externalized into a foundation governed with companies including Cloudflare, Stripe, American Express, Mastercard, and Visa.

### Agent wallets remove the account-by-account payment process
[06:42](https://www.youtube.com/watch?v=YX4Z4XqZ9Vo&t=402s)
Murr says Coinbase has seen around 100 million agentic transactions in the past year, happening on-chain and publicly auditable. The older process required a person to fund each service account, obtain an API key, and give it to an agent. With a wallet or another financial account, an agent can pay each API when it needs the service. Murr compares the old process to buying a separate gift card before shopping at every grocery store. Money gives agents more independence.

### APIs are the underlying layer beneath MCP and CLIs
[08:46](https://www.youtube.com/watch?v=YX4Z4XqZ9Vo&t=526s)
Murr describes APIs as the fundamental access layer. MCP can add context and nuance for natural-language interactions, while a CLI provides a similar interface with a terminal requirement. He expects teams to use all of these forms because agents can wrap the same underlying service in whichever format they can use most easily. He does not treat the current division between them as settled, since the tools are still changing.

### Agents with money could hire services and run longer
[09:29](https://www.youtube.com/watch?v=YX4Z4XqZ9Vo&t=569s)
Murr argues that financial access can let agents continue working without a person intervening at each step. An agent could pay for information on a site that requires payment, work overnight, or hire a lawyer and an accountant while setting up a company. He connects this to the original idea behind a DAO, where automation sits at the center and humans remain at the edges. In his view, agents with money could make that older idea more practical.

### Agent traffic has become a dominant part of Mintlify's usage
[12:24](https://www.youtube.com/watch?v=YX4Z4XqZ9Vo&t=744s)
Lee says agents accounted for about 10% of viewership across Mintlify sites at the beginning of the year and were closer to 80% when the discussion took place. Human viewership was not falling. Agent viewership was growing faster. She says agents resemble developers because they work well with code, Bash, and CLI commands, which makes APIs and other developer-facing interfaces especially important for software companies.

### Technical documentation can be automated after the right context is gathered
[14:04](https://www.youtube.com/watch?v=YX4Z4XqZ9Vo&t=844s)
Lee separates creative prose from technical writing. She thinks human writing remains important for narrative work, while much of the technical documentation process can be automated. That process includes finding related pull requests, Slack discussions, decisions, and other context, then turning it into a concise document and placing it in the right part of a larger content collection. She also admits that people still need to speak with users and teams directly, since agents can hallucinate and lead product teams toward the wrong priorities.

### Teams should count maintenance before building agent tools
[16:41](https://www.youtube.com/watch?v=YX4Z4XqZ9Vo&t=1001s)
Lee identifies build versus buy as a common problem. Teams may build documentation or another productivity system internally, then discover that nobody wants to maintain it. AI makes this problem more visible because the tools and standards change quickly. She recommends focusing company effort on its main value proposition and asking whether the team really wants to own ongoing maintenance. A hackathon prototype is easy to create, but keeping it current is a separate commitment.

## Notable quotes
- Hahnbee Lee: "What ends up mattering a lot more is things like page size, is it taking up a bunch of the context window, and the content being accurate." (01:39)
- Lincoln Murr: "Giving them access to money is how you increase their autonomy." (07:59)
- Lincoln Murr: "I think APIs are the name of the game." (08:28)
- Hahnbee Lee: "At the beginning of the year across all Mintlify sites, we had around 10% of the viewership was agents and now we're closer to like 80%." (12:19)
- Hahnbee Lee: "Hey, we could build it, but do we want to maintain it?" (18:14)

## Tools & references mentioned
- Mintlify
- Coinbase
- GPT-3
- ChatGPT
- x402
- Cloudflare
- Stripe
- American Express
- Mastercard
- Visa
- MCP
- DAO
- Bash
- Fiverr

## Who should watch
- You maintain developer documentation and need it to work for coding agents as well as human readers.
- Your product team is deciding whether to expose an API, MCP interface, or CLI for agent use.
- You are building an internal AI tool and need to think through payment controls, maintenance, and the cost of owning a fast-changing system.

## Related talks

- [The Robots Are Coming for Your Job, and That's Okay](https://aietalks.com/talks/the-robots-are-coming-for-your-job-and-thats-okay) (Elmer Thomas & Maria Bermudez, Twilio, 08:57)
- [Why AI Agents Will Need Their Own Bank Accounts](https://aietalks.com/talks/why-ai-agents-will-need-their-own-bank-accounts) (Lincoln Murr, Coinbase, 16:58)
- [Design like Karpathy is watching](https://aietalks.com/talks/design-like-karpathy-is-watching) (Zeke Sikelianos, Replicate, 19:26)
- [Why AI Agents Still Won't Pay for Anything](https://aietalks.com/talks/why-ai-agents-still-wont-pay-for-anything) (Jan Čurn, Apify, 20:37)
- [Building AI Agents that Actually Automate Knowledge Work](https://aietalks.com/talks/building-ai-agents-that-actually-automate-knowledge-work) (Jerry Liu, LlamaIndex, 17:57)
