
November 23–27, 2026
Senior Software Engineer
Perplexity
As AI agents transition from simple single-prompt completions to complex, multi-step autonomous workflows, the demands on API infrastructure and execution runtimes increase exponentially. Agentic workflows rely on real-time data fetching, streaming responses, parallel tool calls, and prolonged request lifecycles - making traditional backend patterns vulnerable to latency spikes, event loop starvation, and resource burn when downstream LLMs or external tools slow down. IWorking on the Agent API every day, I share practical architectural patterns for building and scaling high-performance infrastructure for AI Agents. Drawing from hands-on experience on Perplexity’s Agent API and large-scale high-load platforms, the session covers: 1) Architecting APIs for multi-step agent loops, streaming execution, and state management. 2) Preventing resource burn by detecting client disconnects early and propagating request cancellations through the entire tool-execution pipeline. 3) Implementing runtime feature controls, rate limiting, and fallback strategies when agent tool chains experience peak traffic or external API downtime
See every session across the AI Agents Conference November 2026 line-up.
View Agenda