Agents in Production: correctness, context, and control
Getting an agent to produce output is the easy part. Keeping it correct as data changes, memory grows, and systems scale is where things get interesting.
This meetup is about what actually holds up in production: fresh context, durable memory, validation, observability, and guardrails that don’t collapse under real-world pressure.
Join engineers from Pydantic, CocoIndex, and SurrealDB as they share practical patterns, hard lessons, and what tends to break first when agents leave the lab.
Bring your hardest questions! We’re here for the honest conversations - the edge cases, the trade-offs, and the things you only learn once something ships.
⚠️ Important – AWS registration required
We’re delighted to be supported by the AWS Builder Loft for this event. Because the meetup is hosted in their space, all attendees are required to register via the official AWS event page for building access — a Luma RSVP on its own is not sufficient.
👉 Required registration (for entry): https://events.builder.aws.com/LyRwZD
Only attendees registered via the AWS link will be admitted by building security on the day.
Agenda
17:00 - Doors + drinks + light bites
18:00 - Welcome
18:05 - Talks
19:30 - Drinks + networking
20:30 - Close
Talks
Samuel Colvin (Founder, Pydantic)
Controlling the wild: from tool calling to computer use
There's a continuum from traditional tool calling through to full computer use, with interesting options at every point along it. This talk is about one particular answer: Monty, a sandboxed Python interpreter built for AI agents. Come watch my code fail in microseconds.
Linghua Jin (Co-Founder, CocoIndex)
Incremental compute engine for reliable AI system
As AI systems become increasingly autonomous, the bottleneck is no longer model capability — it’s keeping context fresh and relevant. Enterprises sit on constantly changing documents, codebases, and multimodal assets where even 1–10% daily drift can break reasoning, retrieval, or agent behavior. CocoIndex introduces an incremental, Rust-powered compute engine built specifically for AI-native workloads. Developers write simple Python transformations without managing deltas, DAGs, or orchestration logic. CocoIndex continuously applies minimal updates to downstream systems — delivering fresh context for AI.
Tobie Morgan Hitchcock (CEO & Co-Founder, SurrealDB)
How to build deterministic agents: from vector search to context layers
Most agent systems don’t fail because the model is bad. They fail because the context is wrong.
Vector search is useful, but it’s only one piece of the puzzle. Once you’re dealing with live systems, changing data, and agents that need to be right more than they need to be clever, “similar enough” stops being good enough.
In this talk, Tobie will show why production agents need a real context layer - not just vectors, but vectors combined with metadata, graph relationships, and structured state. Using a live demo, he’ll walk through what breaks when retrieval is too loose, and how a more complete approach can make agents more accurate, more up to date, and far more dependable in the real world.