Strands Agents with Clare Liguori

The shift toward model-driven agent architecture prioritizes flexibility and reasoning over rigid, brittle workflows. By utilizing system prompts, tool sets, and model selection, developers can build reliable agents that handle unexpected user inputs more effectively. Reliability is maintained through evaluation kits and "steering hooks," which allow for runtime checks and stateful decision-making without the overhead of complex, hard-coded procedural logic. As LLMs continue to improve, the focus is transitioning from simple chat interfaces to long-running agent harnesses capable of managing complex, multi-turn tasks. This approach emphasizes modularity, observability via OpenTelemetry, and the ability to adapt to new models rapidly. Clare Liguori, a senior principal engineer at AWS, highlights that successful agent development requires a scientific mindset, continuous re-evaluation against live traffic, and the willingness to discard outdated scaffolding as model capabilities evolve.