Every single agent will need to verify themselves to prove their intentions on the internet. That's why we're pioneering Web Bot Auth, a protocol that allows for good Agents to use the web. We wrote a blog about the internals and how it works: browserbase.run/wba

One agent to rule them all? @pk_iv of @browserbase doesn’t think so. At the Machine Earning AI Summit, he predicts a growing mix of specialized personal agents, with more competition.

Browser evals are only valuable if they hold up on the live internet. Most evals break with site updates, bot protection, and faulty graders. We partnered with @hud_evals to write about what makes a good browser task that frontier labs actually want. Read the full post: browserbase.run/hud

From Pets to Cattle: Agent Identity on the Web with Keycard, Browserbase, and Visa "Role-based access controls rely on the idea that I know it's Peyton and I trust him, and so he can always go and delete repositories in GitHub. But that doesn't work in an agentic world. We need adaptive access controls, we need adaptive permissions, we need adaptive policy based on the thing. And so that's the big next unlock. What we need is Auto Mode for the web." - @ianlivingstone An agent trying to book one dinner reservation made roughly 200 API calls an hour to Resy and got its user's account banned. At @browserbase Navigate, @vtahowe sat down with @ianlivingstone (CEO, @KeycardAI), @peytoncasper (Product Lead, Identity, @browserbase), and Ivy Lee (VP Product, CMS Agentic Commerce, @Visa) to discuss why the web wasn't built for agents and how agent identity can help. We get into: > Why the web can't tell a good agent from a bad bot > How Web Bot Auth gives browser agents a cryptographic identity tied to a real person or org > How Visa's Trusted Agent Protocol verifies an agent and it's intent to make a payment > Why permissions have to be scoped to the task > What "Auto Mode for the web" looks like TIMESTAMPS (00:00) Pets to cattle: why identity is at the heart of production agents (01:15) Agent Baseline and the ability to respond (02:10) Why you can't revoke access from an agent holding a long-lived API key (03:15) The Resy incident: 200 API calls an hour and a banned account (04:10) Web Bot Auth: identity that travels with agent traffic (06:05) Why Resy blocks agents: good intent vs. bad intent traffic (08:05) Visa's Trusted Agent Protocol and the Visa agent registry (10:10) Good intent, wrong action (11:40) Optimize does not mean delete: permissions scoped to the task (13:10) The Terraform lesson and granular permissions (14:20) Payment tokens and transaction-level controls (15:30) Keeping agents from calling an API 200 times an hour (20:30) Restraint without losing the agent's capability (22:50) Auto Mode for the web (24:00) Visa mandates and the Lakers tickets problem (26:00) AAuth missions and a delegation lesson from Coinbase (27:00) Economic incentive is what drives protocols

Stagehand can use Stripe's new WebMCP tools in checkout pages. Run page.tools() to see what WebMCP connectors a page exposes. With Stripe, you can use Stagehand to add items to cart and complete the checkout with native tools.

Browserbase (@browserbase) is now available in Merge Agent Handler. Agents can fetch a page, search the web, or hand a plain-English task to a browser agent that runs on its own. Check it out: docs.merge.dev/merge-agent-h…

Your Agent needs its own Identity. Join the Coinbase Space today at 12PT (3pm ET) to hear about how agentic payments are driving a need for agents to have delegated identity models.