Building & Governing Banking Agents w/ Arcade.dev & LangSmith Deep Agents

​Getting an agent to work in a demo is the easy part. Getting it through a security review is harder. Who is the agent acting as? What is it allowed to touch? When can the agent delete, modify or share sensitive data? Can you show exactly what it did?

​In this half-day, hands-on workshop, we'll build a loan-processing agent for a bank and take it from first run to a governed, observable deployment.

If you’d like to participate, please apply to attend here:
https://www.arcade.dev/events/toronto-roadshow-workshop-november-3-2026/?utm_source=luma

What to expect:

You’ll deploy the agent with Managed Deep Agents on LangSmith and connect it to email, Slack and custom lending tools through Arcade. Next you’ll add the controls regulated teams ask for:

• ​Just-in-time authorization scoped to each user

• ​Policy checks that block out-of-bounds actions

• ​Human approval for high-value decisions

• ​Redaction of sensitive financial data before it reaches the model

​Finally, you’ll trace and evaluate every step in LangSmith. You’ll take away working artifacts (code base, agent and MCP server) you can reuse, extend and ultimately deploy into production.

​Agenda

• ​9:00 – 9:15 · Joint · Welcome and setup

• ​9:15 – 9:45 · LangChain · Building and deploying agents with Deep Agents and LangSmith, ADLC

• ​9:45 – 10:30 · Arcade · Connecting agents to business systems with just-in-time authorization

• ​10:30 – 10:45 · Break

• ​10:45 – 11:30 · Arcade · Governing agent actions with policy hooks and human approval

• ​11:30 – 12:30 · LangChain · Observing, evaluating, and taking agents to production with LangSmith

• ​12:30 – 1:00 · Joint · Q&A and wrap-up

​Space is limited, so please register to reserve your spot.

​This workshop is designed for engineers, architects and technical leaders who are building or evaluating AI agents. Bring a laptop; setup instructions will be shared with registered attendees ahead of the event.