Clare Liguori

When features that took two weeks now ship in an afternoon, the bottleneck shifts from writing code to making decisions. Frontier teams have discovered this firsthand, achieving 3-10x productivity gains by fundamentally rethinking how developers work with AI agents. This talk covers the practices that separate frontier teams from those who merely "sprinkle" AI on their existing workflows: running agents asynchronously for hours, investing in comprehensive agent steering files, enabling local integration testing for agent self-correction, and automating everything from coding to operations to documentation. You'll learn how teams at Amazon slowed down to speed up, the temporary productivity dips they accepted, and the organizational changes required to sustain this velocity.

Large context windows have become a popular answer to the growing complexity of AI agents. When agents lose track of details, forget prior decisions, or degrade in reasoning quality, the instinct is often to add more tokens. In practice, this rarely fixes the problem and often makes it worse. Bigger context windows increase cost and latency, introduce noise, and amplify failure modes like lost-in-the-middle effects, context collapse, and brittle summarization. This talk argues that the real challenge is not context size, but context engineering. In this session, we will explore practical context engineering techniques for building AI agents that reason reliably over time without relying on ever-larger context windows. Starting from a stateless agent, we will walk through progressively more advanced strategies, including short-term and long-term memory, conversation curation policies, retrieval-augmented generation, and tool-driven context injection. We will examine common failure modes such as context pollution from tool outputs, brevity bias during summarization, and reasoning degradation as conversations grow, and show concrete ways to mitigate them. The talk is grounded in real agent implementations using the Strands Agents SDK and Amazon Bedrock AgentCore, but the principles apply broadly to any agent framework. This session is intended for engineers building AI agents beyond simple chatbots who want practical techniques they can apply immediately. Speakers: Morgan Willis — Amazon Web Services (AWS); Clare Liguori — Amazon Web Services.