Agents SDK

Why Eval++ Is the Next Great Compute Primitive — Sunil Pai & Matt Carey, Cloudflare

Jun 8, 2026 · 24:51

Matt Carey and Sunil Pai from Cloudflare's agents team argue that Durable Objects — stateful serverless with 15ms London latency — and Dynamic Workers — a safe, sandboxed eval for LLM-generated code — form the next compute primitive for AI agents. They explain how Durable Objects enable resumable streaming, multi-tab sync, and background scheduling out of the box without distributed systems engineering. Dynamic Workers allow running generated JavaScript strings in isolated sandboxes, reclaiming 30 years of avoided eval. The pair tease upcoming talks: one on collapsing Cloudflare's 2,600 API endpoints into a 1,000-token MCP tool, and another on a coding agent harness built entirely on Workers that they are already shipping.

OpenAI on Securing Code-Executing AI Agents — Fouad Matin (Codex, Agent Robustness)

Jul 30, 2025 · 14:00

Fouad Matin, an OpenAI engineer working on agent robustness, argues that code-executing agents like Codex CLI require layered security safeguards as they become capable of autonomously writing and running code. He emphasizes three main defenses: sandboxing agents via containerization (e.g., ChatGPT spins up fully isolated containers) or OS-level sandboxing using SeatBelt on macOS or Seccomp and Landlock on Linux; disabling or limiting internet access to prevent prompt injection and data exfiltration, with configurable allowlists for approved endpoints; and requiring human review through PR approvals and monitoring tasks to catch unintended actions like installing malicious packages. Matin also details new API tools—local shell and apply patch—that enable agents to execute code and apply diffs securely, and recommends using remote containers (available in the Agents SDK and Responses API) as the safest deployment option. He concludes by acknowledging that while LLM-based monitors are valuable, deterministic system-level controls remain essential, and invites applications to OpenAI's new Agent Robustness team.

Building Agents (the hard parts!) - Rita Kozlov, Cloudflare

Jul 23, 2025 · 21:12

Rita Kozlov, VP of Product for Cloudflare's developer platform, presents the building blocks of AI agents—client, AI reasoning, workflows, and tools—arguing that effective agents require all four components. She highlights the Model Context Protocol (MCP) as a standard for exposing APIs to LLMs, and demonstrates Cloudflare's Agents SDK, which simplifies hosting remote MCP servers with built-in OAuth, state management via durable objects, and real-time WebSocket communication. Kozlov cites real-world impact: companies using agents for sales automation see 20% revenue increases, 90% faster support response times, and 50–75% time savings. She walks through a human-in-the-loop credit card approval workflow built with Nok, showing how durable objects maintain long-running state, prevent duplicate actions, and route approvals across Slack, email, or in-app notifications. The talk emphasizes that once an MCP server is deployed, it can be used directly from Cursor, Claude, ChatGPT, or a custom client, including voice interfaces via WebRTC-to-WebSocket translation.

Building Effective Voice Agents — Toki Sherbakov + Anoop Kotha, OpenAI

Jul 20, 2025 · 17:17

Toki Sherbakov and Anoop Kotha from OpenAI argue that speech-to-speech models have reached a 'good enough' tipping point for production voice agents, highlighting the shift from chained architectures (transcription + LLM + TTS) to the real-time API's speech-to-speech approach. They detail trade-offs across latency, cost, accuracy, UX, and integrations depending on use case—consumer apps prioritize low latency and expressiveness, while customer service demands accuracy and tool integration. Key design patterns include delegating complex tasks to smarter models like O4 mini, prompting for voice expressiveness and tone, and starting with few tools. For evals, they recommend starting with observability, using SMEs for labeling, then transcription-based LLM-as-judge evals, audio evals with GPT-4 Audio to assess tone and pacing, and synthetic conversations between two real-time clients. Guardrails should run async with a configurable debounce period (e.g., every 100 characters). Examples from Lemonade (early focus on evals and guardrails) and Tinder (customization for brand realism) illustrate successful approaches.