Notte Built Browser Infrastructure for AI Agents, and 100 Companies Are Already Using It

The Macro: Web Automation Has Been Broken for Twenty Years

Every company eventually tries to automate something on the web. Scrape pricing data from a competitor. Fill out forms across ten different portals. Monitor inventory on supplier websites. Log into a bank portal and download statements. The motivation varies, but the outcome is remarkably consistent: it works for a while, then it breaks, then someone has to fix it, then it breaks again.

The reason is simple. Websites are designed for humans, not machines. Every time a site updates its layout, renames a CSS class, adds a CAPTCHA, or changes its authentication flow, every bot that interacts with it breaks. Selenium scripts are famously fragile. Puppeteer and Playwright improved the developer experience but did not solve the fundamental brittleness problem. You are still writing code that targets specific page elements, and those elements change without warning.

The rise of AI agents made this problem urgent. An AI agent that can browse the web, fill out forms, compare prices, and complete transactions is enormously valuable. But building one requires solving not just the AI reasoning problem but also the browser infrastructure problem. You need headless browsers that scale, proxy networks that avoid detection, CAPTCHA solving, credential management, and session handling. Each of these is its own engineering challenge.

Companies like Browserless and Bright Data have built pieces of this infrastructure, but they are tools for developers, not platforms for AI agents. The gap between “I can run a headless browser” and “I can deploy a reliable AI agent that operates on the web at enterprise scale” is enormous. Filling that gap is a billion-dollar opportunity.

The Micro: Two MIT Researchers Who Studied Why Agents Fail

Andrea Pinto (CEO) and Lucas Giordano (CTO) met six years ago and have been working on web automation ever since. Both led ML and LLM research at MIT, with additional stints at Oracle and IBM Research. They did not start Notte because they thought AI agents were cool. They started it because they spent years studying exactly why web agents fail and got tired of watching the same problems go unsolved.

Notte (Y Combinator Summer 2025) is a full-stack browser infrastructure platform built specifically for AI agents. The architecture breaks down into three components: Browser Sessions for instant, stateless browser instances on a global edge network; Browser Agents that handle complex navigation from single prompts; and Browser Functions for serverless logic colocated with browsers.

The numbers are compelling. Under 50 milliseconds of latency. Over 1,000 concurrent browser instances. 99.9 percent uptime. And an agent success rate above 90 percent, which is the stat that matters most. When your agent needs to log into a portal, navigate three pages, fill out a form, and download a file, the difference between 90 percent reliability and 60 percent reliability is the difference between a product and a science project.

The platform includes built-in CAPTCHA solving with antibot detection, residential proxies across a global network, and what they call Agent Vaults with AES-256 encryption for secure credential storage. They even provide real email addresses and SMS numbers for two-factor authentication, which solves one of the most annoying problems in web automation: the verification step that kills most bots.

Pricing starts at $0.05 per browser hour, with a free tier of 100 browser hours. The developer plan is $20 per month plus usage, and the startup plan is $100 per month. They support Playwright, Puppeteer, Selenium, browser-use, Stagehand, and integrations with OpenAI, Anthropic, and LangChain. They are SOC 2 Type II certified, which is unusual for a five-person startup and signals they are serious about enterprise sales.

Over 100 companies are already using the platform, including Fetch.ai and Fingerprint. The team is five people in San Francisco.

One feature worth highlighting: the Anything API. You describe a task in natural language, and Notte converts it into a production API endpoint that returns structured data. So instead of writing a scraper, you write a sentence.

The Verdict

Notte is solving a real problem that gets worse as AI agents become more capable. The better the reasoning gets, the more the bottleneck shifts to browser infrastructure. An agent that can think brilliantly but cannot reliably interact with a website is useless.

The competitive moat here is the combination of infrastructure and intelligence. Browserless gives you browsers. Bright Data gives you proxies. Apify gives you scrapers. Notte gives you all of that plus the AI agent layer, the credential management, the CAPTCHA solving, and the reliability engineering, in a single API. That integration is the product.

The risk is that the big cloud providers eventually build this. If AWS or Cloudflare decides to offer managed browser infrastructure for AI agents, Notte’s five-person team has a fight on its hands. But right now, nobody is offering what they are offering at this level of integration.

Thirty days, I want to see the agent success rate hold above 90 percent as usage scales. Sixty days, I want enterprise contracts, not just developer signups. Ninety days, the question is whether Notte becomes the default infrastructure layer for AI agent companies or whether each agent company builds its own. If Notte wins the infrastructure layer, this is a generational infrastructure company. If every agent startup rolls its own browser stack, the market fragments and nobody wins.