A shared agentic platform for Wood Mackenzie, on Amazon Bedrock AgentCore
Wood Mackenzie built APEX, a shared agentic platform on Amazon Bedrock AgentCore, to eliminate duplicated infrastructure and enable production‑scale agents. The platform centralizes orchestration, safety, observability, identity, and connectivity, addressing the “evaluation and observability” blocker that prevents most proofs‑of‑concept from scaling. By using AgentCore, teams avoid rebuilding authentication, guardrails, memory, and tracing, cutting the 88 % proof‑of‑concept failure rate and allowing rapid rollout of agents like Woody, Lens AI, and the ST Trading App.
⚡ Key Takeaways
- 88 % of AI proofs‑of‑concept never reach widescale deployment at Wood Mackenzie.
- APEX leverages Amazon Bedrock AgentCore as the runtime that provides built‑in orchestration, safety, observability, identity, and connectivity.
- The platform resolves the primary production blocker: lack of reliable evaluation and observability for non‑deterministic agents.
- Teams interact with APEX via APEX Studio, a single control plane for operating agents.
- The solution requires adopting AgentCore; existing self‑hosted stacks would need to be replaced to gain shared infrastructure.
Engineers can move from isolated prototypes to enterprise‑grade agents without re‑implementing core services, dramatically reducing time‑to‑value and the risk of misbehaving agents in production.
✅ Practical Steps
- Deploy Amazon Bedrock AgentCore as the shared runtime for all agent services.
- Use APEX Studio to configure identity‑aware entitlements, guardrails, and observability for each agent.
- Migrate existing agent prototypes (Woody, Lens AI, ST Trading App) onto the APEX platform to consolidate memory, tools, and evaluation pipelines.
Want the full story? Read the original article.
Read on AWS ML Blog ↗