Cloudflare pushes AI agents
Cloudflare expanded its Agent Cloud with tools designed to build and scale long‑running AI agents that can retain state, resume work, and operate across tools and workflows. Coverage notes the update supports proprietary models and offers enterprise customers the option to deploy agents at the edge, and OpenAI release notes reference improvements to GPT‑5.4 context handling and Codex availability on Windows (siliconangle.com) (itbrief.asia) (help.openai.com).
Cloudflare said on April 13 it is expanding Agent Cloud with new tools for AI agents that stay running, keep state, and resume work later. (cloudflare.com) An AI agent is software that does more than answer one prompt: it can call tools, take multi-step actions, and come back to unfinished jobs. Cloudflare said many current setups still rely on always-on virtual servers or isolated sandboxes that are too expensive to scale to “dozens of personal agents” per user. (cloudflare.com) Cloudflare is building this on Durable Objects, its stateful serverless system that combines compute with storage in one place. The company’s documentation says each Durable Object has a globally unique name, durable storage, strong consistency, quick startup, and the ability to shut down when idle. (developers.cloudflare.com) That matters because long-running agents need memory and coordination, not just raw model calls. Cloudflare’s pitch is that developers can keep an agent’s state close to where it runs on Cloudflare’s network instead of stitching together separate servers, databases, and schedulers. (developers.cloudflare.com) (cloudflare.com) Cloudflare said the new Agent Cloud package is meant to move agents from laptop demos to “production-grade workloads” on its global network. Chief executive Matthew Prince said the company has spent nine years building the foundation in Cloudflare Workers and now wants to make Cloudflare “the definitive platform for the agentic web.” (cloudflare.com) The timing lines up with a broader push toward models built for longer jobs. OpenAI said on March 5 that GPT-5.4 supports up to 1 million tokens of context and adds native computer-use capabilities in Codex and the application programming interface, letting agents operate computers and carry out workflows across apps. (openai.com) OpenAI’s model release notes say GPT-5.4 Thinking in ChatGPT improved context window management and better maintains context for questions that require longer thinking. The same notes say the model combines reasoning, coding, and agentic workflows into one system. (help.openai.com) Cloudflare tied its rollout directly to that model wave. In the company’s announcement, OpenAI product lead Rohan Varma said Cloudflare is making it easier for developers to deploy agents powered by GPT-5.4 and Codex for enterprise workloads at scale. (cloudflare.com) OpenAI has also widened where developers can run those coding agents. ChatGPT release notes say the Codex app became available on Windows for plans that include Codex, with support for running multiple agents in parallel, isolated worktrees, and reviewable diffs. (help.openai.com) OpenAI’s developer docs say the Codex app on Windows runs natively with PowerShell and Windows sandbox support, or through Windows Subsystem for Linux 2. That gives the current agent push a second half: not just better models, but more places to deploy and supervise them. (developers.openai.com)
