Amazon and Cloudflare Add Tools to Help AI Agents Browse the Web

Amazon Web Services and Cloudflare have both released new capabilities for building and running AI agents—software that can think through problems, take actions, and carry out tasks on their own. These updates show that the companies are moving beyond simply running AI models and toward building full platforms where agents can work reliably in real business settings.

Amazon added new browser capabilities to its Bedrock AgentCore platform, letting AI agents navigate websites the way humans do—using proxies, saved user profiles, and browser extensions. It also added a dashboard tool to track where AI traffic is coming from across company networks.

Cloudflare took a different approach, focusing on making it easier for developers to build agents. The company released the first remote Model Context Protocol (MCP) server, made some of its infrastructure available for free to agent developers, and launched a general workflow tool designed for agents that need to do multiple steps over time.

What Amazon Is Building

Amazon's Bedrock AgentCore is now ready for companies to use in live business settings. The new browser features are important because many companies still run old software that wasn't built to work with AI—it only has a human-facing website. With these tools, an AI agent can click buttons, fill out forms, and navigate web pages just like a person would. This lets agents handle work that used to require humans to sit in front of a screen.

Amazon also added a way for security teams to see what AI traffic is moving through their networks. This separates AI activity from regular user traffic, which helps companies watch for problems and enforce security rules.

What Cloudflare Is Building

Cloudflare's strategy centers on removing obstacles for developers who want to build agents. The remote MCP server lets developers connect agents to external tools and data without having to set up and manage their own servers. Think of it like giving an agent a phone number to call when it needs help—the agent doesn't need to worry about who answers or how the system behind it works.

The company also made its "Durable Objects" available for free to agent developers. These are basically storage containers that remember information as an agent works through a series of steps. Agents often need to pick up where they left off, and these storage containers help them do that.

Cloudflare also released its Workflows tool for agents, which handles the complicated job of making sure multi-step tasks finish correctly, even if something goes wrong in the middle.

Finally, Cloudflare updated its AI platform so developers can use models from different companies—OpenAI, Anthropic, and others—without having to rewrite their code when they switch.

Easier Development Tools

Both companies made changes to help developers write agent code faster. Cloudflare updated its tools so that when developers use AI-powered coding assistants like ChatGPT or Claude, those assistants understand agent development and can suggest code automatically. Instead of writing everything from scratch, a developer can describe what the agent should do in plain language and get working code back.

What This Pattern Means

This is familiar ground in cloud computing. When Amazon first dominated cloud services, it started with basic storage and computing power. Over time, companies noticed developers were building similar things again and again—apps for phones, containerized software, serverless functions—so Amazon and others built specialized tools for each. AI agents appear to be the next thing worth specializing in.

The detailed focus on browser automation, traffic monitoring, and keeping track of agent state suggests companies aren't just testing agents anymore—they're getting ready to use them for real work. These capabilities address practical concerns that only matter when agents are running actual business processes at scale.

Looking ahead, the parallel infrastructure work from both Amazon and Cloudflare suggests that building agents will follow the same path as previous technology waves. At first, people use general-purpose tools, but as more companies adopt the technology, they demand specialized tools designed specifically for agents.

The emphasis on making development faster and deployment easier also signals that agent development is moving from pure research to practical engineering. When major infrastructure companies optimize for production deployment rather than just making models available, it usually means the technology is ready for real business use.

For organizations considering whether to build AI agents, these infrastructure updates matter. Companies now have clearer, more complete toolkits for production deployment. The availability of browser controls, traffic monitoring, and state management means organizations won't have to build as much custom software themselves.