Agent Infrastructure: Browsers, Control Planes, Orchestration & Security

Cloudflare launches Kitesurf, a cloud-hosted browser for AI agents (beta). Cloudflare launches Kitesurf, a cloud-hosted browser that runs AI agents in Workers-based sandboxes during its beta. This gives practitioners a serverless sandbox pattern for running agents with browser-like capabilities, accelerating safe experimentation and deployment (Principle 07).

Unifying Workers AI and AI Gateway into a single AI control plane. Cloudflare unifies Workers AI and AI Gateway into one control plane to centralize routing, observability, and billing across models and providers. Outcome engineers get a concrete control-plane pattern for model routing and end-to-end observability needed to operate agent fleets in production (Principle 11).

What are Agentic Workflows?. Databricks defines agentic workflows that let autonomous agents plan, adapt, and execute complex enterprise tasks while stressing governance, auditability, and production-grade infrastructure. This frames the architecture and organizational shifts you must design for when turning experimental agents into reliable, auditable delivery lanes (Principle 09).

Menlo Security targets real-time AI agent security with MARS platform. Menlo launches MARS to monitor AI agents at runtime and enforce policies to stop prompt-injection and other live attacks. Implementing runtime policy enforcement becomes a non-negotiable layer for any outcome-oriented agent system to maintain integrity and reduce operational risk (Principles 10 and 14).

Stanford runs 37,000 AI agents as a virtual biotech — one drug design confirmed by Merck. Stanford runs a 37,000-agent Virtual Biotech that designs molecules and gets at least one AI-designed drug independently validated by Merck. This demonstrates that large-scale agent orchestration can produce verifiable real-world outcomes and sets expectations for orchestration, validation, and evidence pipelines in outcome engineering (Principles 09 and 16).