Webinar

Beyond the Context Window: Memory, Forgetting, and 'Dreaming' Between Agent Sessions

October 14, 2026 | 12:00 PM PDT (UTC-7)

Register

Join us for a Complimentary Live Webinar Presented by Merge

As agents move from single-session demos into always-on production systems, the context window becomes the central engineering problem: what should an agent remember, what should it forget, and how does it get smarter between sessions instead of starting from zero? Gil Feig, Co-Founder and CTO of Merge, shares a practical engineering perspective on memory/reflection loops and "dreaming" grounded in real production tradeoffs.

  • What belongs in persistent memory vs. working context?
  • How do you decide what to forget, and what happens when you're wrong?
  • Does offline "dreaming"/reflection actually improve production behavior?
  • Practical context-engineering patterns that survive messy enterprise data
Register
Gil Feig

Co-Founder & CTO, Merge

Speaker

Gil Feig is the co-founder and CTO of Merge, where he builds the infrastructure that lets AI agents securely access and act on real business data across the enterprise. His recent focus is on what production agent systems actually need to hold up at scale, from context and memory design to the realities of running agents against messy, real customer data. Previously, Gil led engineering at Untapped and was a software engineer at Wealthfront and LinkedIn. A graduate of Columbia University, he lives and works in San Francisco.