Strands Agents SDK Adds Steering Handlers for Production-Reliable Agents

Amazon Web Services

Mar 19, 2026 · Updated Apr 25, 2026

The Strands Agents SDK now includes Strands Steering, which intercepts agent behavior at critical moments with just-in-time guidance. In testing across 600 runs, steering handlers hit 100% accuracy — outperforming prompt engineering at 82.5% and graph workflows at 80.8%.

Strands Agents SDK, AWS's open-source framework for model-driven AI agents, adds Strands Steering — Python handlers that intercept execution before tool calls and after model responses. Instead of cramming behavioral rules into prompts that models routinely ignore, handlers inspect the agent's real-time tool call ledger and return Proceed or Guide signals with targeted correction. Handlers are plain Python — deterministic, unit testable, and fully reproducible. For nuanced checks like tone enforcement, they can call an LLM judge.

Across 600 evaluation runs, steering achieved 100% accuracy vs. 82.5% for prompt engineering, 80.8% for graph workflows, and 99.8% for SOPs — while using 66% fewer input tokens than the SOP approach.

Add handlers as plugins to any Strands agent to enforce workflow ordering, parameter constraints, and tone — no prompt edits needed when agent behavior drifts.

View the full update on strandsagents.com

AWS AI

@AWSAIMar 18

Introducing Strands Steering: a new way to build AI agents that stay reliable in production. Instead of patching prompts every time an agent misbehaves, steering handlers intercept the agent at critical moments and give it exactly the guidance it needs.

View on X

Every HeadsUpAI update is written based on its original source and reviewed before it's published. Read our editorial standards →

See all AI news & updates from AWS →

Keep reading

AWS Releases Strands Evals to Systematically Test Non-Deterministic AI Agents

AWS introduced Strands Evals, a framework that uses LLM-based judges and multi-turn simulations to evaluate AI agents. Unlike traditional software testing, this system measures non-deterministic behaviors like helpfulness, tool accuracy, and goal success. It provides a structured path for moving agents from experimental prototypes to reliable production deployments.

OpenRouter Agent SDK Adds Human-in-the-Loop Tools to Pause High-Stakes AI Tasks

OpenRouterMay 8

OpenRouter Agent SDK Adds Human-in-the-Loop Tools to Pause High-Stakes AI Tasks

OpenRouter introduced human-in-the-loop controls to its Agent SDK, allowing developers to build agents that pause for review on sensitive actions. The update includes hooks to auto-resolve routine tasks while surfacing high-stakes decisions to a human interface. This standardizes how autonomous agents ask for help without losing conversation state.

OpenAI Updates Agents SDK to Support Secure Sandboxes and Durable Execution

OpenAIApr 15

OpenAI Updates Agents SDK to Support Secure Sandboxes and Durable Execution

OpenAI updated its Agents SDK with a model-native harness and native sandbox execution to help developers build production-grade autonomous agents. By separating the agent's logic from its execution environment, the system enables durable runs that can recover from crashes and execute code safely in isolated workspaces.

ElevenLabsMar 26

ElevenLabs Launches Guardrails 2.0 for Production Voice Agents

ElevenLabs released Guardrails 2.0 in ElevenAgents, a three-level safety layer that validates user inputs, blocks agent responses in real time, and enforces custom business policies in natural language. For voice agents, a well-crafted system prompt alone isn't enough in production.