What Cloudflare’s Rebuilt Browser Run Actually Is
Cloudflare Browser Run is a managed, headless Chromium service that gives AI agents on-demand browser sessions for automation, testing, and web interaction, now rebuilt on Cloudflare’s Containers platform to deliver higher serverless concurrency, lower API response times, and tighter integration with a full agent infrastructure stack. According to Cloudflare’s Browser Run team, concurrency jumped from 30 to 120 simultaneous browsers, delivering a 4x increase without any required code changes for existing users. Response times for quick actions improved by around 50%, helped by pre-warmed regional browser pools and a new single-request execution path for short tasks. For developers, this turns Browser Run from a shared utility into a core browsing layer that behaves like a first-class edge service, sitting alongside Workers, Sandboxes, and Dynamic Workflows as part of a unified environment for building and running agents end to end.
From Shared BISO Infrastructure to Dedicated Containers
The biggest architectural change in Cloudflare Browser Run is its move off the shared Browser Isolation (BISO) infrastructure onto dedicated Containers. Previously, Browser Run had to coexist with human browsing sessions, where long-lived, steady usage clashed with the spike-heavy patterns of AI agents starting short browser tasks in bursts. The new design creates regional pools of pre-warmed containers, so headless browsers are ready to go when an agent needs them, which directly supports higher serverless concurrency and smoother scaling. State management also shifted from Workers KV to a D1 plus Queues combination, replacing eventual consistency with transactional assignments that can handle up to 500,000 containers per location. That transactional backbone avoids race conditions when many agent tasks compete for browser instances, keeping scheduling predictable even under heavy load. The result is a browsing layer tuned for fast, spiky workloads rather than human browsing habits.
Cutting API Response Times with a New Execution Path
Cloudflare redesigned how quick browser actions execute inside Browser Run to cut API response times. Instead of a multi-step WebSocket choreography, short tasks now run as single HTTP requests executed entirely inside the container. This removes handshake overhead and network round-trips, which is especially important when agents chain many small browser calls as part of a workflow. Combined with pre-warmed Chromium instances, the new path yields roughly 50% faster responses for these quick actions. For developers, this means fewer latency spikes when orchestrating flows like login sequences, DOM checks, or screenshot capture. Faster completion at the browsing layer also reduces the risk of cascading delays across higher layers of the agent infrastructure stack, such as Dynamic Workflows or Agent Memory. In practice, the win is not just raw speed but more predictable timing, which simplifies retry logic and timeout configurations in complex agent systems.
Completing Cloudflare’s Six-Layer Agent Infrastructure Stack
With Browser Run rebuilt on Containers, Cloudflare positions it as the browsing layer in a six-layer agent infrastructure stack that covers compute, orchestration, memory, browsing, and commerce. Dynamic Workers provide isolate-based compute that starts in milliseconds for lightweight tasks, while Sandboxes offer full Linux containers for jobs that need git, bash, dev servers, or multi-language builds, with credentials injected through an egress proxy so agents never see raw tokens. Dynamic Workflows add a roughly 300-line, MIT-licensed orchestration layer, letting workflow code vary per tenant or request and hibernate sleeps for free. Agent Memory, now in private beta, extracts structured memories and retrieves them via five-channel parallel search with Reciprocal Rank Fusion. Browser Run supplies headless Chromium with DevTools and WebMCP support, and a commerce protocol co-designed with Stripe lets agents create accounts, register domains, and deploy. Together, these pieces form a vertically integrated platform for AI agents.
What the New Platform Means for Agent Developers
For developers, the rebuilt Cloudflare Browser Run and the completed agent infrastructure stack reshape how agent workloads are designed and deployed. Higher concurrency—120 simultaneous browsers instead of 30—means teams can safely scale up multi-tenant or multi-agent systems without fragmenting workloads across separate providers. Lower API response times reduce latency budgets for steps involving websites, dashboards, or web apps, which makes complex workflows more responsive. Because Browser Run now sits on the same Containers platform as Sandboxes and connects through the Agents SDK, developers gain a unified environment where compute, browsing, and memory share consistent APIs and deployment patterns. This lowers integration overhead and makes it easier to build capabilities like automated testing, web-based research, or deployment automation into agents. The key shift is that browsing is no longer a bolt-on add-on; it is a first-class layer in an edge-distributed agent platform designed for high concurrency and cost-efficient idling.