Why AI Agents Need Version Control

Key Takeaways

Hi, I'm Filipe. Today I'm going to show you Git Agent on Lize Studio. This feature allows you to add version control to your AI agents and manage separate environments like dev and production. If you build agents in production, you know they evolve fast. You're constantly tweaking prompts, adjusting instructions, and changing models. But without version control, there's no history of what changed, no way to roll back if something breaks, and no way to experiment on a separate branch without losing our stable config. In software engineering, that was solved decades ago with Git. Now we have the same thing for agents. Git Agent in Lize Studio gives you full Git-based version control for any agent. You get environment branches like dev, staging, and production. You get commit history for every change, and you get a direct link to GitHub, so your agent config lives right alongside your code base. Let me show you how it works. I've got a new agent here in Studio. It's a customer support agent for a SaaS product. To enable version control, I scroll down to the bottom where it says version control and enable Git Agent. Doing so will open the configuration window. GitHub is the provider. I'll add my GitHub account info and paste in my GitHub personal access token. You can retrieve yours by going to developer settings under the general settings of your GitHub account. I'm going to leave repository name and working branch set to default and hit save configuration. Now when I create my agent, it will be stored in a new and private repo in the Git Agent standard. If we had to my repositories in GitHub, you can see our agent was created in a new repo, and it follows the agent name in Lize Studio under the main branch. Now let's say I want to test a new behavior for this agent without touching production. I'll create a dev branch. Currently, when a user asks to cancel their subscription, this agent is set to direct them to the support team. On the dev branch, I'm going to change that. I want it to try a retention offer instead, giving users 20% off before they leave. Now, let's test it. I'll say, "I want to cancel my subscription." As you can see, the dev version comes back with, "Before we proceed, I can offer a 20% retention discount on your next quarter." Now, if I switch back to the main branch and give it the same input, "I want to cancel my subscription." Production responds the way it responded before. "I'm sorry to hear you'd like to cancel. I can connect you with our support team." No discount, no experimental behavior. The branches are completely isolated. Both versions are intact. I can go back and forth, test the new approach, and if the retention offer doesn't perform well, my stable version is right here in the main. Nothing was lost. And every change is tracked. In Git history, I can see the full commit history. Who changed what and when, and if something goes wrong, I can roll back to any previous version. There's also a dev mode that lets you see the raw files. Your Soul MD, agent config, including all the settings that define your agent. You can edit these files right here or open them directly in GitHub. And since it's all synced to GitHub, your agent definition lives in a real repo. You can use your existing workflow, CI/CD pipelines, whatever your team already uses. So, that's Git Agent on Lizer Studio. Version control for your agents, separate environments, commit history, GitHub sync, all built in. If you've already seen our Git Agent video on the open standard, this is the same repo format now integrated directly into Studio. I'll drop the link below. Thanks for watching. See you in the next one.