Why Bigger Context Windows Won't Save Your Agent — Elizabeth Fuentes Leone, AWS

A bigger context window won't save your agent. Past a point, it forgets the middle. Elizabeth Fuentes Leone, developer advocate at AWS, explains why agents break when tools return huge amounts of data, and why bigger context windows aren't the fix: models remember the start and end of a long context but lose the middle. Using the open-source, model-agnostic Strands Agents framework, she walks through four strategies (externalize, select, compress, isolate) with code. She covers sliding-window and summarizing conversation managers, short-term, long-term and graph memory, memory pointers that keep big tool outputs out of context, sharing pointers across a swarm of agents, capping tool calls to stop endless loops, and async handles for slow MCP tools. She ends with the anti-patterns to avoid. In this talk: • Why bigger context windows don't fix agents, and how models lose the middle • Four strategies: externalize, select, compress and isolate • Memory pointers, conversation managers and multi-agent state in Strands • Stopping endless tool loops and handling slow MCP tools SPEAKER Elizabeth Fuentes Leone, Developer Advocate, AWS LinkedIn: https://www.linkedin.com/in/lizfue X: https://x.com/lizfue Website: https://elifuentes.tech LINKS Strands Agents: https://strandsagents.com Strands Agents on GitHub: https://github.com/strands-agents CHAPTERS 0:00 Intro 0:37 Why agents break on big tool outputs 1:12 Example: a log-watching agent 1:57 Bigger windows aren't the fix 2:12 Lost in the middle 2:42 What context engineering is 3:17 Externalize, select, compress, isolate 4:07 Conversation managers 5:07 What Strands Agents is 5:57 Summarizing conversation history 6:42 Short-term, long-term and graph memory 8:12 Memory pointers for large tool outputs 9:47 Sharing pointers across a swarm 10:42 Agents that never stop 11:22 Limiting tool calls 12:12 Async handles for slow tools 13:41 Anti-patterns to avoid 14:36 Cheat sheet Recorded at the AI Engineer World's Fair 2026 in San Francisco. Subscribe for more talks from the engineers building with AI. AI Engineer: https://ai.engineer YouTube: https://www.youtube.com/@aiDotEngineer X: https://x.com/aiDotEngineer LinkedIn: https://www.linkedin.com/company/aidotengineer/ #ContextEngineering #AIAgents #AIEngineer