Cloudflare released Kitesurf, a lightweight web browser designed for AI agents that runs in V8 isolates on Cloudflare Workers instead of using Chromium. The browser uses 3.1–7.0× less memory and CPU than Chromium on agent tasks like screenshots and HTML extraction, though it runs 1.7–1.8× slower on wall time. Existing Puppeteer and Playwright clients can use Kitesurf by adding a single parameter, making it cheaper to run agent workloads at scale while maintaining compatibility with current automation tools.
Amazon Bedrock AgentCore gateway now supports rate limiting for AI traffic, allowing organizations to control how much traffic individual users and groups can consume across inference models, agents, and tools. The feature offers three rate-limiting metrics: requests per minute/second (RPS/RPM), tokens per minute (TPM) for inference targets only, and connections per second (CPS) for managing long-lived sessions. Organizations can define granular, multi-layered limits using JWT claims, IAM principals, and target names, enabling scenarios like per-user caps nested within per-group quotas to prevent resource monopolization.
Naïve, a startup offering infrastructure for AI agents to automate business operations, raised $28.5 million in Series A funding led by Nexus Venture Partners. The company has gained over 30,000 developer customers within months and scaled annual run-rate revenue 10x to the low double-digit millions in the past six months. With the new capital, Naïve will develop inference optimization, model routing, memory systems, and serverless runtimes to reduce the cost of running autonomous agents for customers.
Anthropic now recommends running multiple coding agents in parallel using separate git worktrees, but this creates bottlenecks downstream of code generation where infrastructure—staging environments, databases, and runtime systems—still operate as singular shared resources. Faros AI telemetry found that teams with high AI adoption merge 98% more pull requests while review time grows 91%, overwhelming infrastructure not designed for parallel change validation. Teams need to implement branch primitives at every layer of the stack (CI, deployment, database, runtime) to handle multiple concurrent changes, following patterns already established by Vercel, Neon, and Uber's SLATE architecture, so that each change exists end-to-end as a cheap delta rather than queuing behind shared bottlenecks.
Cloudflare has open-sourced Cloudflare OS, a platform that lets organizations deploy AI agents with access to internal systems and tools. The platform internally at Cloudflare since May has been used by thousands of employees across non-engineering functions to create documents, automate tasks, and build apps. The key innovation is a security framework where agents start with no access and resources are mediated through Gatekeepers that enforce fine-grained policies, ensuring data exposure cannot exceed what individual users are authorized to see.
Microsoft's Webwright system converts successful AI agent solutions into reusable, executable code skills that run standalone without models in roughly 40 seconds with zero tokens. Skills are verified twice—first by checking the original solve was correct, then by replaying the distilled code independently—and grow through parameter extraction when multiple verified runs of the same task are aligned. This shifts web automation from repeated model inference to a library of verified programs that can be composed, versioned, and scheduled like traditional software.
Mem0 is a platform designed to provide a persistent memory layer for AI agents, enabling them to retain and recall information across interactions. The system stores context and user preferences to improve agent performance over time without requiring retraining. This allows AI agents to deliver more personalized and contextually aware responses in ongoing conversations.
Microsoft researchers developed SkillOpt, a text-space optimizer that trains natural-language skill documents to improve agent performance while keeping target models frozen. A skill trained on Codex for spreadsheet tasks scored 81.8 on Claude Code, exceeding Claude Code's own in-domain result of 80.4, demonstrating strong cross-harness transfer. Procedural skills like spreadsheet inspection transfer well across models and harnesses, while reasoning-heavy skills remain more tied to their training environment, enabling optimize-once-deploy-everywhere workflows.
Every AI story that matters,
in your inbox by 8am.
TLDRocket reads all relevant sources, removes duplicate coverage, and summarises the
day in two minutes. Follow companies and topics for alerts, or get the
briefing in Slack. Free, no spam, unsubscribe anytime.
Reading TLDRocket needs no cookies, and the readership counts we rely on come from
our own cookieless analytics. Google Analytics is the exception: it sets cookies and
reports to Google, so it stays switched off until you allow it. You can change your
mind any time from “Cookie settings” in the footer.
Strictly necessary
Session security and form protection (tldrocket-session,
XSRF-TOKEN, 2 hours). The site cannot work without them,
so they need no consent.
Always on
Google Analytics 4 (_ga,
_ga_<id>, up to 2 years). Measures which
stories and sections readers use. Google acts as a third-party processor and may
store the data outside the EU. No advertising, no profiling, no data sold.
