I have been through this exact phase and honestly, seeing a kid move past Scratch is pretty exciting because thats when the real engineering starts. Moving into physical hardware was the most satisfying way I found to keep the interest alive without it feeling like kiddie stuff anymore. Since you are in Seattle and have about $80 to play with, you can actually get some professional-grade components that bridge that gap perfectly.

• BBC Micro:bit V2 Go Bundle
• This is the standard starting point. It has a built-in microphone and speaker that works great for basic sound-triggered logic.
• DFRobot HuskyLens AI Machine Vision Sensor
• This sensor is incredible for the price. It handles face recognition, object tracking, and line following without needing a constant PC connection.
• SunFounder Raspberry Pi Pico W Starter Kit
• A great alternative if he wants to dive straight into Python logic instead of block coding. I have been very satisfied with the HuskyLens specifically because it feels like a real tool rather than a toy. It works well with the Micro:bit and the direct feedback on its built-in screen makes the AI logic visible and understandable. I have had no complaints with this setup and it really helps transition from clicking buttons to actually understanding how a machine sees the world. It definitely feels like a solid step up from the basic browser tools... ngl he is gonna love seeing the sensor actually recognize his face for the first time.

Honestly if hes already over Scratch, you need to get him into Python based hardware. It is the best way to bridge that gap. For an $80 budget, I would look at the DFRobot HuskyLens AI Vision Sensor. Its a bit more plug and play than raw sensors but still requires logic to integrate with a controller. If he wants something more advanced, the M5Stack UnitV2 AI Camera Standalone is a solid pick. It runs a Linux system and supports things like face recognition and shape tracking natively. Its much more technical than Teachable Machine because you have to actually handle the data output in code. These kits feel like professional tools instead of toys, which usually helps keep 10 year olds engaged longer. Plus you can find plenty of documentation online for free so you wont need to spend extra on courses.

Jumping in here... I was a bit worried about getting something too complex that would just sit in a box, but we have been really happy with the Makeblock mBot Neo Coding Robot. It fits that $80 budget if you find a good sale and honestly it works well because it uses the CyberPi board. That thing handles voice recognition and some basic AI logic way better than the older stuff. The hardware feels very sturdy and reliable which was important to me because I didnt want to spend hours fixing broken parts. He can still use blocks if he wants but it makes moving to Python really easy once he gets the hang of it. No complaints so far, everything has been pretty smooth. Its a solid way to move past the kiddie stuff without getting totally lost.

Regarding what #4 said about "Jumping in here... I was a bit worried..." - unfortunately, I had issues with those all-in-one robots like the mBot. They often lock you into their own proprietary apps, which feels like a step backward if the goal is actual engineering logic. I found the experience not as good as expected because it still hides too much of the complexity behind their specific ecosystem. Basically, it feels more like a toy and less like a tool. If he is 10 and ready for a challenge, i'd skip the pre-built stuff. Instead, look at something like the Waveshare ESP32-S3-Sense paired with a Freenove Ultimate Starter Kit for ESP32-S3. It is a tiny board with a camera and microphone that lets you do actual TinyML projects using MicroPython. It is much more professional and forces him to actually deal with code and circuitry. Ngl, it might be a steeper learning curve, but thats how you actually learn the logic instead of just clicking buttons in a closed environment. You can get the board and a full kit for way under that $80 limit.

Just saw this thread and I have to chime in because the DIY route is such an amazing way to learn! If he is ready to move beyond basic clicking, I think exploring TinyML is the most professional next step. You can get an Arduino TinyML Kit for right around your $80 budget and it is basically a tiny AI brain in a box. Unlike the simpler kits, this focuses on the actual engineering pipeline of data collection and model training. He can record his own voice commands or motion patterns and then deploy the code directly to the board. It is super satisfying to see a little chip recognize your gestures! It feels much more like a legitimate tech project than a toy, which sounds like exactly what he needs right now. The documentation is fantastic and it really bridges the gap between basic blocks and real-world machine learning.

> Moving into physical hardware was the most satisfying way I found to keep the interest alive. Agreed, hardware makes the theory feel much more tangible. I've been very satisfied using the Yahboom Micro:bit V2 AI Vision Kit for this. It fits your $80 budget perfectly and bridges the gap between simple blocks and Python.

• The vision sensor handles color recognition efficiently.
• It allows for a structured transition to text-based code.
• Great value for the price point.