In this workshop, Bowen Zhang, Member of Technical Staff at Perplexity, will share how Perplexity is building and operating agentic systems that power real workflows at scale. The session will focus on the design choices behind Perplexity Computer, with a close look at how orchestration, model coordination, and task delegation come together in production. Attendees will learn how a modern agent orchestrator can help teams move beyond isolated LLM calls toward reliable, multi-step systems that research, reason, and act. Bowen will also cover the trade-offs involved in building for speed, transparency, and control, and the engineering lessons learned from shipping these systems inside a fast-moving AI environment. - How Perplexity structures agent orchestration in production. - What it takes to coordinate multiple models, tools, and workflows effectively. - Lessons from building systems that need to be fast, interpretable, and dependable. - Practical ideas builders can apply to their own agentic AI stack. This session is best suited to engineers, AI practitioners, and technical leaders who are building or scaling applied AI systems and want a grounded view of what works in production.