Scale AI coding agents while controlling costs with ug – the Unity Gateway CLI

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: Scale AI coding agents while controlling costs with ug – the Unity Gateway CLI · Channel: Databricks · 05:12 · Watch on YouTube

Topics: AI agents, Software engineering with AI, AI governance and regulation, Markets and economy

Summary

Amber demonstrates how an administrator can use the Unity Gateway CLI to configure and distribute governed coding agents across an organization. The configuration sets authorized agents and models, MCP servers, skills, tracing, and budget-based smart routing. Developers apply the configuration through a one-time workspace authentication, and new sessions pick up the latest settings automatically. The demo also shows an agent answering a question using an Atlassian MCP server and skills, with the result and source visible in traces. Unity Gateway provides developers access to coding agents while giving organizations central control over cost, model access, and security.

Key points

Administrators can publish and manage an organization-wide JSON configuration through the Unity Gateway page in Databricks.
The configuration can specify coding agents, authorized models, access levels, MCP servers, skills, tracing, and budget limits.
The demo enables Claude Code and Codex, with smart routing and custom model choices.
Budget intervals route requests to different agents or models: Claude Code for 0–60%, Codex for 60–80%, and a lower-cost model for 80–100%.
The Unity Gateway CLI picks up the latest configuration when a developer starts a new session.
Developers authenticate once to their Databricks workspace to apply the CLI configuration.
The Codex demo uses an Atlassian MCP server to search Confluence for Boba Bricks’ FY26 training goals and returns a cited answer.
Native OTel traces let administrators inspect agent activity, including skill usage and cited outputs.
The presentation concludes that Unity Gateway offers centralized control without requiring developers to change how they work.

Chapters

Start	Chapter	Summary
00:04	Configuring Unity Gateway	Amber introduces the Unity Gateway CLI and begins setting up an organization-wide configuration for governed coding agents.
01:36	Publishing budget-based routing	The active configuration enables agents, models, skills, and MCP servers, with model routing tied to budget intervals.
03:09	Testing Codex and smart routing	A Codex session searches Boba Bricks’ Confluence pages through the Atlassian MCP server and returns a sourced answer.
04:11	Reviewing traces and budgets	Amber inspects the agent’s trace and budget setup, then summarizes Unity Gateway’s organizational controls and availability.

Quotes

I want to talk to you about how I can control coding agents cost at scale using Unity Gateway. — 00:04

The UGC I picks up the latest configuration whenever a developer starts a new session. — 02:07

above 95% training completion is the goal by Q2 — 03:40

The Unity Gateway CLI UG gives developers access to the coding agents they want, while giving organizations central control over cost, model access, and security. — 05:12

Concepts and entities

Unity Gateway CLI, Unity Gateway, Databricks, Claude Code, Codex, GPT-5.6, MCP, Atlassian MCP server, OpenTelemetry, Amber

Full transcript (original language)

00:04 Hello everyone. My name is Amber. I’m on the tech marketing team here at Databricks. And I want to talk to you about how I can control coding agents cost at scale using Unity Gateway. Today, I’ll be an admin at the company Boba Bricks, where my goal is to configure UG, the Unity Gateway CLI for managing and distributing governed coding agents across an enterprise. In my Unity Gateway page in my Databricks workspace, in the right-hand tab under govern, I’m going to click CLI configuration.

00:35 Here, I can distribute and manage govern AI tools for my team. This configuration defines authorized models, access level, and budget limits across our organization. And to implement this, I’m going to click publish configuration, where I can edit the JSON configuration directly. And here we can see everything that I’m able to set. So, this is our default agent, our MCP server, skills, smart defaults, and exactly how they go into the JSON file. And I have a custom

01:05 file already ready, so I’m just going to paste that here. We can see the default agent I have enabled, which is Claude Code. I have smart routing available. I have my default models. I also have tracing enabled. And then I do the same for Codex, so the model services I want. And then if I scroll all the way down to the bottom, I can see my smart defaults. So, this is how I set my budgeting capabilities. And let me show you how

01:36 that looks when I publish it. Now we see the full active configuration. So, again, I have my default as Claude Code. I have the custom models available. I have the default model for Claude Code that I want it to use. I do the same with Codex. I have skills enabled. I have MCPs enabled. All the system MCPs like GitHub, Slack, Google, Atlassian, my smart defaults enabled. So, this is what I set up with those spend intervals. So,

02:07 I say from 0 to 60% use Cloud Code and the models that I have available there. From 60 to 80% use Codex and a specific system model. And then if I reach 80 to 100% of my budget, I want to use a lower cost model like Kimmy. Now, the UGC I picks up the latest configuration whenever a developer starts a new session. So, these updates will roll out without requiring my developers to manually reconfigure their agents. Let me show you what this looks like by copying over the configure

02:38 command. Just going to paste this into my terminal. I have to do a one-time authentication into my Databricks workspace. It’s going to verify the connection. And we can see the configuration’s complete. We have Cloud Code and Codex available, so we can switch back between the two. We have registered MCP servers. We have skills available. So, let’s get

03:09 started with UG Codex. And I’m going to ask, “What are Boba Brick’s FY26 training goals?” Here we can see our Unity Gateway smart router is enabled. So, because we’re using Codex, we’re using the GPT 5.6 model. And it gave a reason for why it routed to the specific model citing that this is open-ended, localization,

03:40 cross-cutting scope. And so, let’s see how it did with the question. So, it explored my skills file. It searched through the Boba Brick’s Confluence pages using the Atlassian MCP server. And that’s actually where Bob Bricks has its FY26 training goals. And then here we see the answer above 95% training completion is the goal by Q2, and gives us a source for where it got this information. All right, going back to my Databricks workspace.

04:11 Next to the CLI configuration, I’m going to go into our traces view. Now, remember these are native OTel traces coming from our coding agents into this unified tracing table, and I’m going to click the latest trace ID. We can see some default language that comes from using Codex as a coding agent. I’m going to scroll down to the outputs, so we can see that my skills were used, and we see the output with the cited

04:41 source. And we can see that indeed this is the Bob Bricks page for the FY26 goals. Lastly, I wanted to show budgets. We saw in the CLI configuration that I was able to add smart defaults with the budget inch budget here, and that was by adding the ID from this budget to the CLI configuration in the JSON file. So, in conclusion, the Unity Gateway CLI UG gives developers access to the coding

05:12 agents they want, while giving organizations central control over cost, model access, and security. I’m able to roll out coding agents across our development teams without forcing developers to change how they work. And I’m happy to announce that UG is available today, so I can’t wait for you to check it out. Thank you.