Agents Unboxed: Stories Behind the Build
Join Datadog as we host Letta and Gumloop for an evening of engineer-led lightning talks and practical lessons on building AI agents in the real world. We’ll share what’s working (and what isn’t), patterns we’ve repeated across products, and the tools + workflows we use to ship agents that are reliable, observable, and maintainable.
Expect candid stories, concrete takeaways, and plenty of time to meet other builders.
What you’ll learn
How teams are designing agent architectures (tools, memory, planning, evaluation)
Reliability patterns: offline and online evaluation
Monitoring + observability for agents in production
Common failure modes and how engineers debug them
“From demo to deployment”: what changes when you ship
Schedule
5:30 PM – Networking (Food & beverages provided)
6:00 PM – Talks
6:45 PM – Networking
Speakers
Nicolas Hivon – Datadog
Datadog is building its Natural Language Capabilities to allow users to query logs and metrics in English. In this talk, Nicolas Hivon will discuss the learnings about developing a 'Reasoning Mode' using an agentic flow with tools, its impact on query resolution, and the strategic next steps for improving NLQ capabilities.
Sarah Wooders – Letta
Letta Code is a memory-first coding agent harness, designed to adapt and learn with use. In this talk, we'll cover how Letta Code's memory system has evolved from its original MemGPT-style memory system (based on the 2023 research paper) to its current design, based on the latest advancements in coding agent harnesses and post-training of models like Opus 4.6 and GPT-5.3-Codex.
Gonzalo Soto – Gumloop
Gumloop is building no-code workflows and AI agents used by teams across leading technology companies such as Shopify, Instacart, DoorDash, and Ramp. In this talk, Gonzalo will share the evolution of agents — from early embedded capabilities inside automation tools to standalone products — along with lessons learned along the way. He’ll discuss the origins of the idea, shifts in product thinking, and broader reflections on building agent-driven systems in production environments.
Who should attend
Engineers, product builders, and technical founders who are building with LLMs - especially anyone working on agentic systems, internal copilots, workflow automation, or production LLM apps.
Notes
Space is limited - please RSVP.
Bring questions, demos, and war stories.
About the Hosts
Datadog is the monitoring and observability platform for cloud-scale applications, helping engineering teams see across their entire stack. Datadog LLM Observability is the AI engineering platform to develop, evaluate, and monitor Agents with confidence.
Gumloop is a no-code platform that lets non-technical teams build and deploy AI agents and workflows — no engineering required.
Letta is the platform for building stateful AI agents with long-term memory, founded by the creators of MemGPT out of UC Berkeley's Sky Computing Lab.