Anchor Browser

𝗝𝗲𝘄𝘀 𝗶𝗻 𝗙𝗶𝗻𝘁𝗲𝗰𝗵 - 𝗥𝗼𝗼𝗳𝘁𝗼𝗽 𝗛𝗮𝗽𝗽𝘆 𝗛𝗼𝘂𝗿 Excited that Anchor Browser is teaming up with Yossi Farro and the Jews in Fintech, a community bringing together Jewish professionals in finance and tech. The event is happening on Monday, July 20, 6:00–9:00 PM in our hometown NYC. If you're interested in joining the first JiF gathering, join the waitlist now so you're first in line if a seat frees up. https://luma.com/9ogxoojc Big thanks to the JiF community for putting this together. We look forward to seeing it grow! ⚓⚓⚓

𝗡𝗲𝘄 𝗗𝗼𝗰𝘀 𝗟𝗜𝗩𝗘 No matter how many enterprises adopt Anchor Browser, we will always remember our dev friendly roots. Today we’re excited to share the latest in that mission of dev friendliness with our revamped docs! Take a look at them to see how using our platform can help your agents act on LLM reasoning within unmodified web environments. When you dive in, you’ll notice new documentation that cover two core capabilities to create a seamless system: The first is Agentic Browser Control, which allows an agent to receive a task in plain English and carry it out directly within a cloud-based browser. This allows it to authenticate, navigate and interact with a page as a human user would, even on sites with no automation support. The second is Automation Tasks. You define a task by writing a prompt or demonstrating it, and Anchor compiles it into deterministic code and a reusable workflow. Runs repeat reliably and produce consistent output. Thanks to Noa Rapoport and Yuval Assa for their contributions to this work! This release marks another step toward establishing Anchor Browser as the enterprise-ready browser infrastructure for agents. Read more here: https://lnkd.in/ePt_PJJP ⚓⚓⚓

𝗧𝗵𝗲 𝗘𝘃𝗼𝗹𝘂𝘁𝗶𝗼𝗻 𝗼𝗳 𝗕𝗿𝗼𝘄𝘀𝗲𝗿 𝗔𝘂𝘁𝗼𝗺𝗮𝘁𝗶𝗼𝗻: 𝗙𝗿𝗼𝗺 𝗦𝗲𝗹𝗲𝗻𝗶𝘂𝗺 𝘁𝗼 𝗔𝗜 𝗔𝗴𝗲𝗻𝘁𝘀 Browser automation has changed three times over the past two decades: from scripts that click, to scripts that talk directly to the browser engine. Now, we have agents that look at a screen and decide what to do. What has changed is how automation thinks. What has remained constant is the infrastructure it depends on. Here's our timeline: - 2004 - Selenium/W3C WebDriver - programs that executed predefined clicks and failed when the page structure changed. - 2017 - 2020 Puppeteer/Playwright - faster and more reliable but still reliant on a pre-written sequence of steps. - 2026 - AI agents: given a goal, the model interprets a screenshot and determines the next action independently. Every era still requires an active browser process. The AI driving the agent’s decision does not replace this infrastructure; it is simply a new consumer of it. The driver might be Selenium script, a Playwright script, or a model issuing computer-use actions. Either way, something still has to launch the browser and sustain the session. Only the mechanism for deciding what to click has changed. As agents take over the decision-making, the infrastructure underneath becomes the real differentiator. It’s not whether the agent can click the right button. It’s whether you can run thousands of these sessions reliably while also keeping them authenticated and giving the model a stable environment to act in. That’s the layer Anchor Browser is built for: managed, authenticated cloud browser sessions that any script or agent can connect to.

𝗬𝘂𝘁𝗼𝗿𝗶 𝗦𝘁𝗼𝗽𝗽𝗲𝗱 𝗙𝗶𝗴𝗵𝘁𝗶𝗻𝗴 𝗕𝗿𝗼𝘄𝘀𝗲𝗿𝘀. 𝗡𝗼𝘄 𝗧𝗵𝗲𝘆 𝗝𝘂𝘀𝘁 𝗦𝗵𝗶𝗽 𝗠𝗼𝗱𝗲𝗹𝘀. Yutori is a 14-person team building ‘Navigator’. These are a series of computer-use models that take a screenshot as input and output the next action (click, type, scroll) to complete a task on a website. Instead of training in simulated environments, Yutori trains these models on real, live websites so they don’t fail when they meet the real internet. But every dropped session, proxy failure or bot-detection block during training invalidated the whole RL rollout. Lost rollouts meant wasted compute and fewer usable examples per training run. Engineers were spending less time on actual model development as they were debugging and overseeing browser infrastructure. So Yutori split responsibilities: Anchor Browser handles browser infrastructure, so that Yutori can focus entirely on the model layer. That split gave Yutori three things: engineering time back for model development, a clear indication of whether failure is infrastructure or model mistake. It also made their costs predictable, since their enterprise API bundles inference, orchestration and browser costs together. We're proud to be the infrastructure layer that lets Yutori focus fully on building great models. Take a read through of Gabi Weinberg ⚓'s conversation with Yutori founding engineer Lawrence Chen to see in more detail how Yutori leverages Anchor. https://lnkd.in/gyERabJu

NYC Builders where are you? We'll be here... will you?

What does it look like when your AI co-founder actually pulls its weight? Pancake is building that. And we're powering the infrastructure behind it. The teams building in this space are moving fast. We're glad to be under the hood. What would you build with an AI co-founder?

GPT-5.4's computer-use agent is now live in the Anchor Browser playground. Go try it. This matters more than a typical model update. GPT-5.4 is OpenAI's first general-purpose model with native computer-use capabilities. This means it doesn't route through APIs to interact with software. It navigates interfaces directly. Clicks, inputs, scrolls. The way a person would. And you know we excel in enabling that. This update should increase the efficacy of browser agents in production. With Anchor Browser, that means one line to spin up a secure, isolated browser session and GPT-5.4 handles reasoning steps with confidence. Anchor is the best stage in the world for computer-use models to act on. Playground link: https://lnkd.in/dm6eJfBZ

The Anchor Browser US team is heading to AI Dev SF 26. Last time we were in SF, we launched #OmniConnect and connected with teams in robotics, voice AI, and enterprise, each one moving fast on Computer Use Agents. This trip, we're expecting more of the same. If you're building with agents and want to talk browser infrastructure, find us there. Idan Raman ⚓️, Gabi Weinberg ⚓, Nadav Magnezi still have a few open slots. Worst case, you find them at Sabra. See you in SF.

OmniConnect is live. Read why we built it. ⬇️ Or have your AI developer one-shot it into production today. anchor-skill.md ftw Either way, you'll get out of doing enterprise-grade browser tasks.