AI Agents That Actually Do Work (2 Real Examples)

This Udemy deals page makes me over $1,000 a month without any manual intervention. How? Well, I have an assistant that manages it. An AI assistant. Did you know that AI agents can do more than just intrusively swim through your emails and calendar events and update your popular CEO self at 8:00 a.m. every morning? This agent combs through the hundreds of variants of Udemy deals for me, finds the relevant ones for the upcoming days and weeks, updates my data based on its findings, updates my Udemy page, removes any old stale deals as well, and creates a pull request, tagging me in it on a regular daily basis. Again, it acts as an assistant with only one key task. Like I could pay someone for their time to do this, some percentage of the thousand dollars a month that is made, or I could do it myself, which I've done for quite some time, and it's cumbersome. But instead, we can create an agent dedicated specifically to this. Here's another good example. So, a user has found a significant bug on one of my open- source applications and is currently creating a GitHub issue about it. And they submit the issue, but not to me. Instead, it goes to my AI assistant to do the dirty work of finding the solution. It looks at the issue, investigates the codebase, spins up a container of the site if needed for testing, and figures out the problem. And when it does, it creates a PR for me with the solution committed with a thorough description of the problem and what the fix it proposed does. But I get an alert, and at that point, I can decide to submit the fix or not, just review the PR and merge it in. Agents can be delegated to do some really useful work and companies and individuals that are keeping up with the latest and greatest are already kneedeep in agent assistance. Just look at OpenClaw. There are many people really innovating here and coming up with great use cases for this personal AI assistant. But note the word personal. What happens when you start to fire off multiple or many AI agents to different destinations that have different environments? This is a challenge that many companies are facing. Agent management at scale. I mean, anyone can open up 10 clawed code terminals and fire off tasks. But how do you track progress in events or provide proper environment dependencies for various code bases, run them in the cloud on defined schedules or react to system events? Well, for this you need some sort of orchestration platform. Well, today I'm not only going to walk you through exactly how I created both of these agents, the triage agent and the deal updator agent, but I'm going to show you how to manage these and many others using a brand new orchestration platform. form I came across called Oz from the team over at warp. Now Warp has been one of my main tools for promptdriven development. Warp gives you this entire agentic terminal ecosystem with your choice of models, cloud syncing, easy plug-and-play, MCP, auto suggestions, and now they've introduced a full-blown cloud orchestration platform to manage the agents you're spinning up. These Oz agents are cloud connected background agents that run from events, schedules, or integrations, giving teams scalable automation with shared observability and centralized configs. And if you want to go try it out, I've partnered with Warp on this video. And while Oz is free, they will give you 2500 Oz credits for only five bucks with the promo code Travis or the link that I've shared below. So be sure to go check that out. So how Oz works at a high level is this. So there's a trigger. So think a tag in Slack, an API call, a cron job that kicks off an agent, which can be a skill or a prompt. So a trigger kicks off an agent that runs on an environment. So think GitHub repo or a Docker image. And it creates artifacts like a pull request, a plan, or a session sharing link. And don't worry, I'll show you all about this pathway in the video as we work through recreating these two examples. First, we're going to look at a scenario where a passive agent gets triggered from an event or in our case when a GitHub issue is created. And then second, we'll move on to looking at doing this from the new Oz web interface. So, I've created this simple booking app for this demo with a few bugs in it for demonstration purposes. Now, imagine as I launch my app and I continue building it, more bugs are uncovered and users will inevitably create issues around those bugs. So it would be helpful to have an AI agent there to be the middleman to take some of the load off in some way. So for this demo, I'll create a GitHub action that runs when the label Oz agent is added to an issue. So think maybe your QA team is creating issues, you're a dev and you're looking through new ones and when you see one that you know an LLM can handle, just add the O agent tag to trigger the Oz agent to go figure out the solution and leave a PR with it in it. That's how I'll demo this, but of course it can be any GitHub event you want. So, here's the action I created. And by the way, there are examples provided by Warp to get you started. I actually took the autofix example and then tweaked it to do more of what I wanted to do in this demo. So, looking at my action, essentially, it checks out the repo, dynamically builds a prompt from the issues title and description, hands it off to the warp's Oz agent to analyze the code and make fixes. So, you see here it runs with the prompt in a Warp API key. Then it automatically creates a branch, commits the changes, opens a PR with the agent summary, and comments back on the issue with a link to the PR. All triggered by just adding a label. Again, there are other examples here of reviewing PRs, responding to comments, and then other integrations with Slack and Linear. So, let's go ahead and create an issue that we're having with the app. I can't log in with default credentials. And then for the description, when I try to log in with the demo credentials provided, it fails with the error message of invalid email or password. Imagine someone else is filling this out about a bug they found in my app. So, let's submit the new issue. And then when we add the label, it will kick off the agent. Let's go over to actions and see what's underway. It checks out the repo. It runs the Oz agent. And you can see the issue title and description being passed with instructions for the agent to perform tasks. This can actually be a separate skill which we'll see next. But for this example, we're including it directly for the Oz agent in the GitHub action. And you'll see the agent working through the request, reading the relevant repo files as it locates the problem. It uncovered the issue and it creates a linked PR for the fix. That is very cool. Let's go back and see the pull request. So here it describes what I found, what file, what the issue was, and of course updated the file to fix our error. And do note that this doesn't have to be created in GitHub. Maybe it's a form in your web app that users can submit issues in the Oz API triggers the cloud agent. There's lots of flexibility here in utilizing these new cloud agents. Let's move on to the second example. So, not only can you use GitHub and other options to trigger an agent, Warp also provides a web app for centralized management and monitoring of runs that are kicked off from the cloud. So, automations, API or SDK calls, schedules, etc. And of course, if agents are kicked off locally in your warp terminal, the conversation's on your machine and there's no cloud visibility. But if it is a cloud session, then all activity shows up in the web app. And the appeal here is that you can close your laptop, you can walk away, and let it work. and then if you need to check in, you can pull it up on your phone while on the go. Let me show you how this works in the context of my Udemy Deal Hunter agent that I talked about at the beginning of this video. Let's get this set up. So, first in the Oz web app, I'll add GitHub as an integration, and I'll let it access only this project repo. Second, let's create an environment. Now, this is a core concept in Oz. An environment defines how an agent runs, not what it does. It's the runtime layer, Docker image, repos to clone, setup commands, and the repeatable context every run gets. The environment ensures your cloud agents run with the same tool chain and setup every time, regardless of where they're triggered from. So for this, I need one or more GitHub repos. The agent will clone a publicly accessible Docker image that can build or run your project. And Oz has a number of pre-built ones you can choose from or let it suggest based on your codebase. And then there's setup commands to install dependencies or prep the repo if needed. I don't need any of those. So you have your environment here and how your agent runs. And then it's time to create an agent that can run in this environment. Or a better way to think of this is creating a skill. With this skill, you get consistent behavior, repeatable automation, and sharable workflows. This is the brain of the agent. And I can define this here, but what I like to do is just define it in my codebase as a skill and then Oz will pick it up automatically. So to revisit the Oz flow, there's a trigger. We'll do this manually until we get it working right and then I'll show you how to place it on a schedule that kicks off an agent. This agent is the skill we created that runs on an environment. The app is a NextJS app, so it will need the Docker image to run that and creates artifacts. The artifacts are the updated deals YAML file and updated Udemy page both via APR which I specified. Now the final step is to define a few secrets. If using the CLI I run Oz secret list, you'll see I have three secrets. Two for the platform where I get the deals and then the GitHub token which is a personal access token I created to give lease privilege access to this agent. To add secrets, you run the Oz secret create commands as seen here in the docs. These secrets are stored securely and cannot be retrieved once they're created. And then at runtime, Warp sets the secrets it needs as environment variables for each cloud agent run based off of who triggered the agent and how it was triggered. Mine are personal, so scoped to me. And again, they're never readable after creation. And before we run this, let's look at the skill itself. So I have the skill in my repo that lays out the steps the agent needs to take to complete this run. It fetches fresh data. It reads the deals file. It filters for the month that I ask for, groups deals by campaign, generates deal IDs, converts timestamps, cleans up old deals, updates the YAML file of the deals, updates the actual web page, outputs a table summary, and puts it all in a PR. I can run this on a schedule, but for now, I'll trigger it manually. So, choose new run, select my Deal Hunter agent or skill, select my website environment, and that's it. Let's run it. And you can view the run real time here on the website. Here's my current run among all of my past runs. And the status is running. And check this. If I want to see exactly what it's doing, I can choose view session and I can watch the session run on the web or I can watch it here in my local terminal. So as it wraps up, here's the summary, what it removed, kept, added, and skipped. Now let's go and check the PR it created. And there's the PR. A summary of what it did. and it didn't tag me. So, I need to troubleshoot that. But I can now choose to merge it or not. And going back to the web app, let's go to new schedule, choose their agent, and set it to run daily so we don't miss any deals because they do release deals at the most random times. Very cool. Note again that you can also trigger this via the Oz CLI, the API, or the SDK. And now that I have this environment, any other agents I want to create to do work on my website, I just create a new skill and then reuse this environment. and note some of the other examples they give of an SEO AEO audit, accessibility audits, Terraform style checks, MCP builder, and more. So again, I wanted to share this with you all. I think it's a great new addition to warp, and I think it's a good example of how to use agents in a way that's really useful. And again, while it's free to try out, for a limited time, use the code Travis to get 2500 Oz credits for only five bucks. If you found this video helpful, give it a thumbs up, consider subscribing, and I'll see you in the next video.

Try Warp for free today ➞ https://oz.dev/travisyt AI agents aren’t just for chatting, they can run real business workflows. In this video, I show two real, working examples: First, a GitHub triage agent that reacts to a new issue (or label), investigates the repo, and opens a pull request with a fix. Second, a deal-updater agent that keeps a revenue-generating page on my site fresh by fetching new data, cleaning stale entries, updating files, and creating a PR...and I can run it manually or on a daily schedule. You’ll also see how cloud agent orchestration works (triggers → agents/skills → environments → artifacts) with a new agent orchestration platform from Warp called Oz. Thank you Warp for partnering with me on this video. 🕒 Timestamps 00:00 Dealhunter agent example 1 00:53 Triage agent example 2 01:28 Agents are useful 02:10 Orchestration platform 03:11 How AI cloud agents work 03:52 Triage Agent workflow 06:32 Dealhunter Agent workflow 11:21 Conclusion 📢 Video mentions Try Warp for free today ➞ https://oz.dev/travisyt Oz Agent GitHub examples - https://github.com/warpdotdev/oz-agent-action/tree/main/examples Oz Agent documentation - https://docs.warp.dev/agent-platform Triage agent GitHub repo - https://github.com/rodgtr1/warp-test 🎥 Watch These Next 🎥 https://youtu.be/F3j_1AEQkHk https://youtu.be/pXCRkyLhXDY https://youtu.be/F3j_1AEQkHk FOLLOW ME ON Twitter - https://x.com/travisdotmedia LinkedIn - https://linkedin.com/in/travisdotmedia FAVORITE TOOLS AND APPS: Udemy deals, updated regularly - https://travis.media/udemy ZeroToMastery - https://geni.us/AbMxjrX Camera - https://amzn.to/3LOUFZV Lens - https://amzn.to/4fyadP0 Microphone - https://amzn.to/3sAwyrH ** My Coding Blueprints ** Learn to Code Web Developer Blueprint - https://geni.us/HoswN2 AWS/Python Blueprint - https://geni.us/yGlFaRe - FREE Both FREE in the Travis Media Community - https://imposterdevs.com FREE EBOOKS 📘 https://travis.media/ebooks #aiagents #warp #aiprogramming