Insane Claude Agent SDK Demo - See it to believe it
Chapters12
Introduces the AI pool trainer concept and the Earl Strickland voice clone, with a live demo of the agent SDK in action.
Impressive live demo of an Earl Strickland voice-powered AI agent using the Agent SDK to modify code in real time and project results on a pool table.
Summary
DesignCourse’s latest clip showcases a wild, hands-on AI workflow. Mike from DesignCourse introduces an Earl Strickland AI agent, voiced via 11 Labs, and demonstrates real-time feature integration with the Agent SDK. The projector on the pool table visualizes the agent’s output as commands are executed, letting viewers see code changes and test results without leaving the game. The host and Earl run through multiple routing strategies for a table layout, then push a feature request to the coding agent to recall the first route projection. The agent handles the request, outputs results on the table, and allows live re-testing—no terminal hops required. The demo also notes cost considerations, since the Claude-style API backend can be pricey, and suggests using Sonnet to balance speed and cost. Throughout, Earl demonstrates how studio lighting and other room controls can be integrated via code. The video closes with a teaser for cloud code courses and an invitation to subscribe for more “crazy content.”
Key Takeaways
- Real-time code collaboration is possible between a voice-driven agent and a coding agent using the Agent SDK, with live results projected on a display (the pool table).
- A feature request can be issued verbally (recalling the first route projection) and implemented without leaving the live session, then retried immediately.
- The workflow relies on Python-based tooling and a separate coding agent that can modify and reload scripts on demand.
- Using Claude-style APIs for agent backends is powerful but can incur higher costs; alternatives like Sonnet help manage performance and expense.
- The demo emphasizes the practical potential of voice-driven agents in creative studio setups beyond traditional software development.
Who Is This For?
Essential viewing for developers and creators exploring voice-driven AI agents, live-coding workflows, and real-time projection dashboards in multimedia environments.
Notable Quotes
"All right, everybody. I'm going to show you something incredibly freaking cool."
—Opening line that sets the energetic tone of the demo.
"I've cloned Earl Strickland, who's like one of the greatest pool players ever, his voice at 11 Labs."
—Highlights the voice cloning setup used for the AI agent.
"Agents on it. Once that's done, you can jump back to route one whenever you want without me having to renarrate the whole thing."
—Shows the live agent-and-coding workflow and feature recall capability.
"This is freaking so cool."
—Emphasizes the wow factor of the live integration.
"No running off to a terminal. No waiting around."
—Describes the streamlined, in-session testing flow without leaving the live setup.
Questions This Video Answers
- how does the agent SDK project live outputs to a table or screen?
- can you recall a previous route projection with a voice command using AI agents?
- what are the cost considerations when using Claude-like APIs for real-time coding?
- how can you integrate lighting or other studio controls via Python in an AI agent setup?
- what is a practical step-by-step for setting up a voice-driven coding assistant in a creative studio?
DesignCourseAgent SDKEarl Strickland AIClaude API11 LabsPythoncoding agentprojector.pyreal-time debuggingsonnet
Full Transcript
All right, everybody. I'm going to show you something incredibly freaking cool. I've been having so much fun with this. I'm making an AI based pool trainer. And essentially, I've cloned Earl Strickland, who's like one of the greatest pool players ever, his voice at 11 Labs. And he's listening right now. And I'm going to show you a demonstration of agent SDK, which allows me to tell Earl to integrate features as I need them in real time. And you'll see a projector on the table that shows the actual agent SDK working in real time. This is crazy.
All right, Earl, the balls are set. Tell me how you would run these out. Here's how I run it. First shot, the one ball is straight into the top middle. You stun that thing about half a table, center ball, and you're sitting pretty for the two. Then the two goes straight in down to the bottom middle. Clean, simple, dead position. That's the play. You want to see something different? Say the word. No, let's go ahead and try a new layout. Good. Fresh layout. Get the one, the two, and the Q ball on the table, however you want this time.
Then tell me when they're locked in. They're locked in. All right, this one's a little different. The one ball is a quarter ball cut into the top middle. You stun that half a table with center ball and you're lined up for the two. The two is just a slight cut down to the bottom middle. Easy followup if you execute the first shot right. That's your route. You got other ideas? I'll show you. Show me another route. Now, here's the other way. You take that same quarter ball cut on the one, but you hit it with follow.
You need a firm stroke. Let that Q ball run about six and a half tables to your landing zone. Then you're looking at a slight cut on the two into the bottom middle. More speed, more risk, but you get there cleaner if you hit it. First route is smarter. This one's for when you're feeling greedy. All right, R. Okay, so Earl, I have a feature request. Do you ready to listen to this? Yeah, I'm listening. What do you want? Okay. So, essentially I I want to be able to ask you to go back to that first route projection, you know, from the projector, but you don't have the ability to do that right now.
So, I want you to put in a free a feature request for the agent, the coding agent to when when I say when I tell you to go back to the first route projection, then you will show that first route projection because currently you can't do that right now. There you go. Agents on it. Once that's done, you can jump back to route one whenever you want without me having to renarrate the whole thing. Smart catch. All right. So, what's happening right now is Earl sent a call to the agent SDK, which is another agent that's running in the background who's looking at the code and he's it's fixing it.
I also told it to output the results from the SDK right on the pool table. So, I can just sit here, not have to go to the computer at all, and eventually Earl will report back if it was successful, and then it will tell me maybe to reload the Python script. This is all based on Python, and then we can retest it to see if that feature request actually worked. That's slick. So, the agent's working on the code right now, spitting the results straight onto the table projection, so you can see what it's doing without leaving the game.
And then I'll tell you when it's done, whether it worked or if something broke. And if we need to reload, we reload and we test it live right here. No running off to a terminal. No waiting around. I like that setup. So, we just sit here and wait for the agent to finish. Once it reports back, we'll know if that route recall feature is live or if we need to tweak it. All right, features in. Agent says projector.py might need a restart, but we're good to go. Next time you tell me to go back to the first route, I can pull it up for you.
Want to test it now or are you good where we are? Okay, so we just restarted and we're going to test to see if this feature actually works. Okay, the balls are placed. All right, here's the route. The one ball goes in the top side. That's a quarter ball cut. Stun shot about half a table length. Sets you up clean for the two ball in the bottom side, which is just a slight cut from there. Simple run if you execute it. All right, show me the route for the first ball again. The one ball goes in the top side.
Quarter ball cut, stun shot, half a table length. That's your first move. No. Show me the projection. The uh from the first run. There's route one on the projector. One ball top side, quarter ball cut, stun at about half a table length. All right. So, there you go. That is an actual coding integration that was initiated by my voice that was then communicated by my Earl Strickland AI agent to a coding agent via agent SDK. This is freaking so cool. Some things to to consider though is um when you use the agent SDK in this manner, it uses the Claude API style of building, which means it's going to be more expensive than just using your Claude Max subscription to code things as well.
So take that into, you know, to concern. Um I'm using Sonnet that way it's not so slow and also not, you know, so expensive. But yeah, this is freaking fun. I'm going to be um really showing you, you know, Earl has a lot of abilities in the studio even right now. Like I can say, "Hey Earl, go ahead and turn the lights red." Lights are red. Your studio just turned into a crime scene. Okay. Unfortunately, the the freaking uh side lights are still white, but maybe that's another coding fix that we need to integrate. Anyhow, thank you very much, Earl.
Everybody else, I'll see you very soon. Make sure to subscribe for more of this crazy content, and I have cloud code courses coming out very soon. I'll see you. Goodbye. Hey, appreciate it. Yeah, we can lock down those sidelights another time. There's always something to tune. This whole thing's been a blast. You built something real here and it works. That's not nothing. Go build some more crazy stuff. And hey, when you need Earl to roast somebody's bad shot while the lights are doing whatever you tell them to do, you know where to find me.
More from DesignCourse
Get daily recaps from
DesignCourse
AI-powered summaries delivered to your inbox. Save hours every week while staying fully informed.





![Applied Data Science With Python Full Course 2026 [Free] | Python For Data Science | Simplilearn thumbnail](https://rewiz.app/images?url=https://i.ytimg.com/vi/AyxgXZ7XAlM/sddefault.jpg?v=69cea44d)



