OpenAI/Blog/26/02/27/Stateful Runtime Environment for Agents in Amazon Bedrock
E2B (vendor product)
- Commercial AI sandbox / agent runtime platform (E2B | The Enterprise AI Agent Cloud): short‑lived isolated environments (positioned as microVM‑backed, e.g. Firecracker) for running untrusted or LLM‑generated code, browsers, desktops for computer‑use agents, notebooks, batch RL experiments, and similar agent tool workloads—not a general “rent a VPS” or “deploy my monolith API” product by default.
- BYOC / self‑host: E2B markets deployment in a customer’s own cloud (including AWS accounts / VPC). That path is adjacent to “build on AWS primitives yourself” but still productized around the sandbox API and templates, not around simplifying long‑running ECS services.
- Market sector vs typical AWS compute stories
- Same neighborhood (cloud compute, containers, isolation) as AWS/Lightsail and Amazon ECS offerings such as AWS/ECS/Express Mode in the sense that all ultimately run workloads on AWS (or can).
- Different primary job‑to‑be‑done than Lightsail (simple VPS / managed app building blocks) and than ECS Express Mode (opinionated production web/API on Fargate with auto‑wired ALB, DNS, scaling): E2B optimizes for high churn, many ephemeral sandboxes, agent SDK ergonomics, and template‑driven images rather than steady‑state service hosting.
- Open‑source oriented company with public cookbooks and examples on GitHub under the
e2b-dev org (e.g. github.com/e2b-dev). - Adjacent AWS agent story (different layer): OpenAI and Amazon announced a Stateful Runtime Environment for Agents in Amazon Bedrock—Bedrock‑native persistent orchestration and state for multi‑step agents, not a substitute for sandbox compute. See OpenAI/Blog/26/02/27/Stateful Runtime Environment for Agents in Amazon Bedrock.
- Vs GitHub/Codespaces: both offer cloud environments where code runs, but Codespaces optimizes the human (or single‑workspace agent) dev loop—GitHub repo, DevContainer, IDE—while E2B optimizes many short‑lived, API‑created sandboxes for untrusted / LLM‑generated execution. Teams sometimes use Codespaces to host coding agents (see that page’s long‑running Claude/Code notes); E2B targets productized sandbox scale for end‑user or multi‑tenant agent tools, not replacing the primary Codespaces use case.