OpenAI Agents SDK Update: Native Sandbox, Memory, Snapshotting (April 15)

OpenAI released a significant update to its Agents SDK on April 15, 2026, adding capabilities designed for enterprise-scale agent deployments. Key new features: (1) Native sandbox execution — agents can run code in isolated compute environments without external infrastructure setup; (2) Configurable memory — agents can persist state across sessions with configurable retention policies; (3) Standardized integrations — prebuilt connectors for common enterprise data sources; (4) Portable workspaces — agent state can be migrated across environments; (5) Built-in snapshotting — durable checkpointing for long-running agent runs. Simultaneously, OpenAI shipped GPT-5.3-Codex (unified coding agent model, ~25% faster than prior Codex variants) and the ChatGPT Pro plan ($100/month, unlimited GPT-5.4 + 10x Codex usage).

For GTM engineers building agent-based outbound, research, or attribution workflows: the snapshotting + memory features are the most significant — they make multi-step autonomous workflows durable. Before this update, a failed step in a long agent run (e.g., 3-hour prospect research task) meant restarting from scratch. With snapshotting, you can resume from the last checkpoint. For enterprise GTM teams: the native sandbox and standardized integrations reduce the engineering overhead of building compliant, auditable agent deployments. This brings OpenAI's agent infra closer to parity with Anthropic's Managed Agents (launched April 10).

• "OpenAI's Agents SDK now supports snapshotting. This is the feature that makes 3-hour autonomous workflows practical."
• "Anthropic Managed Agents (April 10) vs. OpenAI Agents SDK (April 15): two different bets on the agentic infrastructure layer."
• "The Agents SDK update puts memory and checkpointing in every agent. Here's what GTM operators can now build."