Hope you enjoy this episode!
As an aged father of little ones, I can’t write AND podcast at the same time. My sleep-deprived brain just can’t handle it.
So yes, the written version below is, in fact, AI generated based on what I discuss on the pod. Thus, the best experience I think is the audio, but in case you can’t experience that, a written version is below.
Let me know if you enjoy it by subscribing below, I hope you do! See you next week…
Slides are also here for visual reference.
Hey everyone, and welcome back to the Robotic Scholar Podcast. Week 51. We’re so close to Christmas and New Year’s. I’m excited; things are slowing down, everyone is getting ready for the holidays, and I honestly really like this time of year. It’s a beautiful time where meetings get canceled and people are taking stock. I hope y’all are enjoying some downtime as well. It’s been a year for me, that’s for sure, and I’m appreciating the break.
The 3D Printer Ejection
One thing that has been making me upset, though, is that I am still in 3D Printer Hell. I thought I got out of it—I even posted on Substack Notes that I finally got the printer working—but the errors just keep coming. I’ve decided to just eject on this printer.
I put the situation through Gemini and asked about the Creality K2 Plus relative to the printer I probably should have bought from the beginning: the Bambu Lab A1 Combo. People seem to love Bambu Lab; they’re like the Apple of the printer space, while Creality is like Android. I have nothing against Android—I love the OS and how it works natively with Google—but I do not want to be training to be a 3D printer mechanic.
I got into this to print robots and shells for my Jetson Orin Nano; I’m not looking to get deep into the mechanical process and swapping out parts. Every time I fix something, something else breaks. Now it’s an MCU error. I don’t even know what that is. I found a Bambu Lab on Amazon—it’ll be here Saturday—and I’m paying someone on TaskRabbit to get rid of this hundred-pound Creality machine Friday afternoon.
The Creality did do some great work, though. I have this R2-D2 leg shell it printed for the robot I’m building. So, shout out to the K2 Plus for that, but we are done.
Gemini 3 Fast & The Platform Advantage
Let’s get into AI news. I am thrilled that they shipped Gemini 3 Fast this week. Their timing was impeccable because ChatGPT 5.2 had just released. I was starting to miss that “polish.” The voice mode on Gemini was leaving a lot to be desired, and being in “thinking mode” all the time was getting old because queries took two or three times longer.
I got Gemini 3 Fast on my phone Wednesday morning. It’s great. I’m not a ChatGPT hater, but the power of these solutions is in the AI plus the platform combination. Gemini clearly got the memo internally to use Google Tasks for storing tasks. It works natively with the products you already use.
I have a hard time seeing how ChatGPT competes with that without building their own email, tasks, and notes apps. You need deep integration to provide the experience Gemini offers out of the box. OpenAI is trying to keep it lean, which is smart in the long run, but how long can you do that while Google is nipping at your heels?
Riding the Frontier Wave
Regarding the product I work on, Assist at Procore: we just did a foundational model upgrade to the latest frontier model. There’s a great article called The Bitter Lesson that talks about how it is so much better to ride the wave of frontier models than to think you can engineer your way into better responses.
Because the new model has been iterating in the background and becoming more agentic, it was night and day at selecting tools and understanding context. We didn’t have to do much; we just upgraded the model. If you’re working on an AI product, it is critical to stay on the frontier. You are better off riding that wave of capability than thinking you can engineer your own capability better. You can’t.
AI Won’t Save a Bad Strategy
The last point is that AI with a poor product or business strategy can’t save you. You can bolt AI onto a product, but if that company is set up to fail in this new world, it won’t work.
The online learning industry is a great example. Before AI, you learned from books and 60-hour video courses. Then AI happened, and the utility of that went away. Why watch hours of video when you can just ask a chat interface your question about Python and get the answer? That business strategy was destined to be bulldozed.
It’s like when Excel shipped. The “bean counter” jobs went away, but analyst jobs rose up. If you aren’t willing to completely recast your product strategy to be additive with AI, you are doomed to fail—slowly, but ultimately. No technology in history has saved bad product decisions. Look at Google: they realized they hadn’t executed on chat, so they changed their whole strategy, reinvested, and now they are arguably one of the best providers of AI models and platforms. You can’t just bolt it on and hope. You have to put the thought in.
On that note, I hope you have a great rest of your week. I’ll do one more episode before Christmas. Enjoy the soft week.
