Browser Use moves down the stack to build an execution layer for AI agents

Why it matters

Browser agents are becoming a real workload, but their economics depend on browser session cost, cold-start latency and blocking risk. Browser Use is trying to compete below the application layer, where infrastructure choices can change whether agent workflows are viable at scale.

In a two-post thread on X, Browser Use (@browser_use), the agent-browser startup, said it has launched a new browser infrastructure layer for AI agents. The company priced access at $0.02 per hour and claimed it is more than 3x cheaper than alternatives.

https://x.com/browser_use/status/2061520845621907564

The more consequential signal is how far down the stack Browser Use says it is going. This is not just hosted Chromium or a browser automation feature. The company described a stack built on a Chromium fork, a Firecracker fork and a custom Linux kernel, with "unlimited scaling" and "subsecond cold starts." Those performance claims were not backed in the thread by benchmarks, test conditions or a named comparison set.

That positioning makes the launch look less like a SaaS feature and more like an attempt to build an execution layer for agents. Firecracker became known as the microVM technology AWS built for Lambda and Fargate to enable fast startup times. If Browser Use is forking at that layer, it is optimizing below where most application startups operate.

In a follow-up exchange, Reformedot (@reformedot) told Tony Ennis it now maintains its own Firecracker fork, saying the change lets it own the stack "end-to-end" for custom low-level optimizations and new features. Reformedot also said it still uses Unikraft for agents and other services.

The product page, browser-use.com/browser, is positioned as something users can try directly or hand to an agent. That framing matters because AI systems that need to open sites, click, log in, scrape or transact through a real browser are constrained by latency, cost per session and blocking risk. At large scale, fleets of autonomous browser sessions can turn commodity cloud and browser stacks into a cost problem.

Browser Use is also signaling that detection resistance is part of the product category, not a side feature. As anti-bot systems and browser fingerprinting evolve, agent companies may find they cannot rely indefinitely on generic browser infrastructure. The company's "stealth browser" framing points to the same conclusion as its kernel and Firecracker work: agent workloads are pushing startups deeper into infrastructure.

The missing details remain material. Browser Use did not state how the $0.02 per hour price is metered, what concurrency limits apply despite the "unlimited" claim, or which incumbent prices it used for the 3x comparison. The announcement establishes the company's bet: lowering the cost structure for browser agents will require control over the browser, VM and operating system layers, not just a better automation API.