I presented at DevelopersIO 2026 Tokyo with the title "Hosting AI Agents on DGX Spark and Collaborating via Slack"
This page has been translated by machine translation. View original
Introduction
Hello, I'm Shimada from Classmethod's Manufacturing Business Technology Department.
At the mini-session of "DevelopersIO 2026 Tokyo" held on October 24, 2026, I gave a presentation titled "Keeping an AI Agent Resident on DGX Spark and Collaborating via Slack."
Presentation Materials
The animated version is available at the following link.
What I Talked About
I framed the question as "What should you decide first when operating an AI agent that handles internal data?" and answered it using the agent running on our team's DGX Spark as an example.
The answer is "what is allowed to go outside, and what you can't afford to have broken."
- What you can ask the agent from Slack (Backlog issue management and task brainstorming)
- A configuration where, for each request, NeMo Switchyard and a judge decide whether a lightweight in-house model or a heavy cloud model responds
- Permission design that allows communication to external services at the granularity of host, method, and path
- An example where things worked as configured but didn't turn out as expected
Closing
The content covered in the presentation is summarized in detail in the following series.
If you're interested, please take a look.
There are also articles comparing results by swapping out the judge.
