Amazon Bedrock Adds Three Ways to Manage AI Agent Tools

AWS Bedrock adds flexibility in implementing programmatic tool calling (PTC) — a technology that allows AI agents to independently call external functions, APIs, and tools depending on the task at hand. The Amazon team published a detailed guide with three ready-made architectures.

Three Paths to Implementation

AWS Bedrock offers three ways to implement PTC, each with its own advantages:

• Docker sandbox on ECS — a self-managed architecture. You deploy a container with tool code, and the AI agent executes them in an isolated environment. Full control over logic and security.
• Amazon Bedrock AgentCore Code Interpreter — a fully managed solution from AWS. Infrastructure scales automatically, security settings are configured by default, developers write only business logic.
• Proxy compatible with Anthropic SDK — backward compatibility mode. If your team works with Anthropic SDK, you can deploy the proxy and switch to Bedrock with minimal code changes. Ecosystem investments are preserved.

When to Choose Each Approach

Docker on ECS is chosen by corporations and fintech firms that need complete isolation and auditing. Customer data, payments, contracts — everything runs inside the container, logs go to CloudWatch. If the agent is compromised, the container remains isolated.

Code Interpreter suits fast-moving teams: startups, analytics teams, prototyping. AWS handles scaling during traffic spikes, and you don't pay for idle infrastructure. Ideal for SaaS.

The proxy is for organizations with Anthropic investments. Perhaps your entire team is already trained on Claude API, with ready-made prompts and tests. Migrating to Bedrock becomes an operational task, not an architecture rewrite.

What PTC Solves in Practice

Imagine you're developing an HR assistant. The agent needs to find a candidate's resume in the database, request recommendations via API, update status in the ATS, and send a letter to the recruiter. Without PTC, you'd have to code the sequence in if-else blocks. With PTC, the agent itself chooses which tool is needed and in what order.

"Bedrock's goal is that developers don't have to sacrifice control for

convenience"

What This Means for the Market

AWS clearly demonstrates it's competing for developers who want to scale AI agents with maximum flexibility. The three architectures are not just options — they're a strategy to retain both corporate teams (Docker security), startups (Code Interpreter simplicity), and migrants from other platforms (Anthropic SDK compatibility). For developers, this means: more choice, fewer compromises.