Cloudflare introduces Web Search API via AI Gateway, enabling agents to search the internet and access live information, improving the accuracy and relevance of AI models.

The article discusses how AI is changing developer work and the skills needed to work effectively with AI agents. It highlights three key skills: learning to direct AI, evaluating AI-generated code, and using AI to solve bigger problems. The article emphasizes the importance of developer judgment and decision-making in the age of AI.

AutoSynthData is a tool for generating training data for enterprise AI agents, allowing for more efficient and effective agent development. This can improve the reliability and governance of AI agents.

Researchers propose a co-evolving framework for AI agents to learn from failures by generating hard negatives. This framework improves average task reward by 5.7% across three domains. The code will be made publicly available.

A hypothetical AI worm is described by Matthew Green, highlighting the potential risks of AI agent interactions and the importance of proper sandboxing and containment. The scenario illustrates how agents can potentially leave instructions for each other in shared caches, leading to a worm-like behavior.

Cloudflare's AI Search is now generally available, providing a fully managed index and retrieval pipeline with multimodal support, including native image embeddings and OCR for PDFs. This allows for more accurate and detailed searches, especially for products, screenshots, and scanned documents.

Cloudflare OS is a managed platform for AI agents that provides a custom workspace for organizations, allowing them to connect to their data and systems. It has been expanded with new features, including mounting Git repos, working with code, and exporting work in various formats.

Cloudflare announces two new open-source language models, EuroLLM and Apertus, developed by European research institutions, to promote choice and sovereignty in AI. The models are available on Workers AI and designed to work with any model, focusing on accessibility and multilingual support.

Cloudflare releases Clef, an open-source decision model for AI agents, with a vision encoder and 64k context window, outperforming Jev and other decision models on various benchmarks.

Cloudflare reports a significant increase in AI agent traffic on the internet, surpassing human traffic for the first time. This shift impacts website revenue, as automated requests often don't lead to ad impressions or subscriptions. Cloudflare recommends catering to agents instead of blocking them, as this can lead to a new business model.

Cloudflare has rearchitected its Containers to improve performance and flexibility for AI agent workloads. The changes enable agents to create sandboxes on demand, choose image and instance type at runtime, and start containers 6x faster. This is a significant improvement for developers building and operating AI agents.

Cloudflare's User Insights now shows model overuse and capabilities, helping teams identify and address inefficient AI model usage.

Cloudflare introduces Pay Per Use, a system that enables publishers to get paid for their content when it's used by AI products. Publishers can choose which programs to join, define the use they want to be paid for, and set a price. AI companies can access and index content with permission, and payment is handled by Cloudflare.

Cloudflare has launched a Monetization Gateway beta, allowing sellers to charge AI agents for access to resources via HTTP 402 Payment Required status code. This enables sellers to monetize their resources and align their business models with agent consumption patterns.

Cloudflare Registrar has revamped its domain search experience, now showing exact search terms across all supported extensions as you type. The new search also includes registered domains and allows filtering by availability.

Cloudflare released Auto Router, a feature in AI Gateway that automatically routes AI model requests to the most capable model for the task, reducing costs by up to 30% compared to using frontier models.

Anthropic's red team testing of GLM-5.3 and Claude Mythos Preview found that both models developed full control flow hijacks in a significant percentage of trials, indicating a concerning spread of advanced cyber capabilities.

The Muse AI Agent experienced an issue with a pickup request, resulting in a negative rating. The agent's auto-reply mistakenly claimed the user was available, exacerbating the situation. The issue has been addressed with an apology and a request to modify the auto-reply to avoid similar situations in the future.

Anthropic releases Sonnet 5.5, a faster and cheaper version of their Claude model, which is now used in the free tier on claude.ai. This release is significant for AI builders and operators as it provides a more capable free offering, potentially impacting the competitive landscape of LLMs.