You can now clone any YouTube channel’s thumbnail style—and automate the process of generating thumbnails for your own videos in that same style.
In this sneak peek, I show how I used the Agent Development Kit (ADK) to replicate Alex Hormozi’s exact thumbnail look using OpenAI’s new image model, gpt-image-1.
Here’s how it works:
• Scrapes thumbnails from a target channel
• Analyzes key visual elements and creates a style guide
• Generates prompts based on your content idea
• Produces thumbnails that match the brand using gpt-image-1
All I had to do was:
• Upload a picture of me
• Include two logos (OpenAI + ADK)
• Write a 2 sentences about the video I wanted to create
The agent handled the rest.
This is just a teaser. The full tutorial drops later this week. Make sure you’re subscribed on YouTube so you don’t miss it.
What other ADK workflows or use cases would you love to see?
watch video on watch page
2 - 0
Everyone’s Talking About DeepSeek and o3-Mini... But Should You Use Them? Instead of following the hype, let’s look at the data.
Here’s what stands out:
🚀 DeepSeek-R1 has completely replaced Claude 3.5 Sonnet.
Same SWE-Bench performance, but at a fraction of the price. If you’re still using Claude, you’re paying extra for no real reason.
💰 Claude 3.5 Sonnet is insanely expensive.
At $15 per million output tokens, it only makes sense for very small coding tasks—otherwise, DeepSeek-R1 is just better.
⚡ o3-mini is the best choice for serious AI tasks.
• 200K token context window
• 1,637 tokens per second throughput (fastest on this list)
• Relatively low cost compared to Claude & GPT-4o
For complex AI applications, o3-mini is a no-brainer.
I’m making the switch—o3-mini for everything going forward.
What’s your pick?
63 - 12
Battle of the AI Models: Which One is Best for Agents?
I’m deep in testing O3-Mini, GPT-4o, DeepSeek-R1, and DeepSeek-V3 to figure out which model is truly the best for AI agents.
To find the winner, I’m stress-testing them against the most common tasks agents need to handle:
🧠 Instruction Overload: Can it handle rule-heavy tasks without getting confused or hallucinating?
🛠️ Tool Call Hell: Can it handle 5 consecutive tool calls, feeding results from one into the next without breaking?
🔍 Needle in a Haystack: Can it retrieve precise information from large datasets while staying contextually aware?
So far, one model is dominating—and I’ll be switching all of my agents over to it going forward.
I’ll share the full results (and the winning model) in my upcoming video—recording tomorrow, dropping Thursday! Stay tuned.
Which model do you think will win? Drop your guesses below! 👇
29 - 9
In the middle of stress testing o3-mini and DeepSeek-r1.
o3-mini is winner - it's not even a competition!
I'm testing each model against the most common developer tasks:
1️⃣ Build a new project from scratch
2️⃣ Build a new feature in an existing app
3️⃣ Refactor existing code and generate tests
-
With o3-mini and cursor, I was able to build a ChatGPT replica that allows me to chat with local LLM models in a single shot and everything works like a charm!
On the other hand, DeepSeek-r1 got stuck and only generated a single javascript file. Far from a fully functional website.
Here's the GitHub repo with the prompts so you can replicate the results on your own:
github.com/bhancockio/o3-mini-vs-deepseek-r1
The screenshot below shows the functional app that o3-mini was able to create in a single shot!
I'm still working on the second 2 tasks but right now, there is a clear winner.
I'll keep you posted as I keep testing.
Also, I'll be putting all of this into a YouTube video that will hopefully come out on Monday!
If you have any questions around o3-mini or deepseek-r1, let me know!
I feel like I have a PHD on these models after all the testing I've done over the past 24 hours.
49 - 9
Hey guys! 👋 I'm super excited to share something that's going to change how you interact with your CrewAI setups.
We've just launched a game-changing feature called `crewai chat` that lets you have real conversations with your crews!
In my upcoming video (dropping early this week!), I'll show you exactly how I used this feature to create a professional newsletter in just 5 minutes. Yes, you heard that right - FIVE minutes! 🚀
Here's the magic: I simply chatted with my crew, refining the output until it was perfect. The result? A newsletter that's ready to send to my community.
Check out the actual newsletter my crew generated on Substack in the link below.
Stay tuned for the video release, and in the meantime, let me know what you think of this new feature!
Would you use crewai chat in your workflow?
Cheers,
Brandon Hancock 👨💻🥂
open.substack.com/pub/aiwithbrandon/p/4-ways-to-cr…
21 - 3
Quick @crewAIInc and LLM Tip:
LLMs are lazy and struggle with word count.
So, if you're creating a Newsletter or Blog Crew where word count matters, give your agent a Word Count Tool (code below 👇).
Without a Word Count Tool, LLM's give you 800 words max by default because they can't count.
With a Word Count Tool, the agent sees it's not generating enough content and continually rewrites the post until it hits the perfect word count!
-
I bring this up because I'm working on a new YouTube video for y'all that shows you how you can chat with your crews using the new `crewai chat` feature.
In this video, we're going to chat with our Newsletter Crew and continually provide feedback until the Crew generates the perfect newsletter!
Normally, writing a newsletter could take hours.
When you're chatting with a Newsletter Crew, it takes a couple minutes!
I'm super excited for y'all to see this video when it comes out early next week!
45 - 5
Starting a YouTube channel as an AI developer has been one of the most rewarding decisions of my career.
Here’s how it’s helped me:
💼 Opened many new job opportunities that helped me 2x my total compensation in just two years.
🎯 Allowed me to create courses and coach others, earning tens of thousands of dollars.
🤝 Helped me connect with amazing people like you, build credibility, and expand my network.
And here’s how it could help you:
🚀 Land better jobs with your channel acting as an online résumé.
🎓 Kickstart a career in freelancing or AI consulting by showcasing your portfolio.
💡 Test and validate app ideas with your audience before committing months to development.
I’m working on a video to show you exactly how to do this, but I need your help choosing the title:
🛠️ Option 1: "How to Start an AI Personal Brand"
🎥 Option 2: "How to Start a YouTube Channel as an AI Developer"
The content will cover the same step-by-step process to help you build your audience, attract opportunities, and grow as an AI developer. I just want to know which title resonates more with you!
If you have any other favorites, please drop in the comments 👇
(And if you have questions about starting your channel, let me know—I’d love to help!)
7 - 13
I'm wrapping up recording for the Sora masterclass right now, but I wanted to go ahead and share a GPT I created for you guys to help you make AMAZING Sora prompts.
This GPT will help you level up your prompt game from this:
"A beautiful woman sitting outside at a cafe."
To this:
"A beautiful woman sits gracefully outside a quaint cafe in a bustling city at sunset. She sips her coffee while gazing thoughtfully into the distance, surrounded by warm, golden light. The setting is alive with soft chatter and the clinking of glasses, framed by lush greenery and bistro tables. Style: Romantic and cinematic."
In the comments below, I've a link to the GPT so you can try it yourself!
16 - 1
Progress Update: Supabase Vector Store Crash Course
Hey everyone!
I’ve got some cool stuff in the works that I think you’re going to love. 🎥
Right now, I’m putting together a video that’ll show you how to set up a vector database for free using Supabase’s Postgres vector DB. I’ll walk you through how to build the vector store, save data with embeddings, and make queries to pull relevant info.
This video is part of a bigger series I’ve been working on. In the next one, I’ll teach AI developers how to build an email autoresponder so they can go sell it to their clients. 📨
Here’s the plan:
✨ Monitor client emails
✨ Store them in a vector DB
✨ Automatically generate draft replies
It’s going to be a game-changer for anyone looking to add AI-driven automation to their services. 🚀
The vector DB crash course is dropping soon, so keep an eye out for it! If you’ve got any questions or ideas for future videos, hit me up in the comments.
Cheers,
Brandon 🧑💻🥂
P.S.
Be on the look out for this Supabase Vector Store Crash Course coming out later this week!
50 - 6
Hey everyone! 👋
Today, I’m super excited to share a sneak peek of my latest video, where I walk you through the ultimate AI tech stack for building MVPs (Minimum Viable Products) lightning fast. If you’ve been trying to get your AI projects off the ground or want to quickly showcase proof-of-concepts to your clients, you’re in for a treat.
🚀 Why This Stack?
In the upcoming video, I break down a powerful combination of Claude, V0, and Cursor. These tools, when used together, help you build beautiful and functional Next.js MVPs in record time.
The Stack Breakdown:
- Claude – Acts as the brains of the operation, helping you brainstorm ideas, outline, and strategize. Think of Claude as your project’s foundation.
- V0 – This is your MVP powerhouse. It can crank out Next.js components like a pro and streamline server-side logic to get your app running fast.
- Cursor – Cursor polishes the details, handles data cleanup, and makes those last-minute tweaks that turn a good MVP into a great one.
✨ Real-World Use Case: Real Estate Offer Letter Generator
Let me show you a real-world example. In the video, I walk through how I built a Real Estate Offer Letter Generator in just a couple of hours:
- Input client details and upload documents like conversations and specs.
- Auto-generate offer letters based on uploaded info.
- Tweak values like offer price, export the document, and send to clients with a few clicks.
The coolest part? What usually takes hours was done in minutes using this stack! 😲
👷♂️ Building the Pyramid: From Concept to MVP in No Time
I like to think of it as constructing a pyramid:
- Base Layer: Claude pulls all your ideas and turns them into a concrete plan.
- Middle Layer: V0 builds out the core components, pages, and server-side logic.
- Top Layer: Cursor polishes the details and handles tweaks to ensure everything runs smoothly.
🎯 Why This Stack Matters for You
If you’re a startup founder, this stack lets you get feedback fast. Instead of spending weeks on a product, you can validate concepts in days. 💡
For AI freelancers, this is your secret weapon to wow clients. Deliver a functioning proof-of-concept in just hours—you’re not just selling ideas, you’re showing what’s possible.
Watch the Sneak Peek 🎬
In the video, I show exactly how I built a fully functional app in just a couple of hours. You’ll see the workflow in action—from brainstorming to finished MVP—so you can start building projects faster using the same approach.
watch video on watch page
0 - 0
I teach developers how to build fullstack AI applications 🤖
4 May 2022