Suggestions
Brad Neuberg
Staff Machine Learning Engineer at Planet
Brad Neuberg is a Staff Machine Learning Engineer at Planet, where he focuses on applying deep learning techniques to geospatial artificial intelligence (AI) problems. His work involves leveraging a fleet of Low Earth Orbit cubesats that capture imagery for various applications, including environmental monitoring and agricultural analysis.23
Professional Background
Neuberg has over 15 years of experience in software engineering, having held significant positions at major companies such as Google and Dropbox. His previous roles include:
- Senior Machine Learning Engineer at Dropbox: He led efforts in developing advanced computer vision systems that processed billions of images using deep learning frameworks like TensorFlow and PyTorch.1
- Research Scientist at the SETI Institute and NASA's Frontier Development Lab: Here, he applied deep learning to space science, focusing on projects like creating high-fidelity virtual telescopes for solar observations.12
Education and Skills
Neuberg holds a degree from Columbia University and possesses expertise in areas such as image processing, computer vision, and advanced machine learning techniques.43 His innovative contributions include developing open-source projects aimed at enhancing satellite data analysis and personal photo organization through deep learning models.1
Notable Achievements
- Co-founded the coworking movement, which has grown to encompass over 15,000 coworking spaces globally.
- Contributed to significant projects that have advanced the deployment of collaborative tools and educational technologies.1
Overall, Brad Neuberg's career reflects a strong commitment to integrating machine learning with practical applications that enhance understanding and interaction with both digital and physical environments.
Highlights
The definition of the Singularity is that we can't make predictions beyond it. It does not necessarily follow that AI replaces everyone after the Singularity, it could be some strange, unexpected fusion or semi-stable state thats hard to predict right now, with humans in the loop. That's the whole point of calling it a Singularity.
There's a few things I'm seeing with AI coding I want to talk about, where AI velocity is being chosen over quality, simplicity, and human comprehension.
First, every team & company gets to decide how they want to work, so these are just my opinions. Also, I'm very pro AI, so this is just nuance when it comes to how and where we use these tools. Also, I expect these opinions to be revisited in 6 months to a year based on where things are at (strong opinions loosely held).
I actually do believe we are in a slow takeoff of the Singularity, so take this with a grain of salt as this is my own personal report as I grapple with how to use these rapidly changing tools and my own role in context with them.
In my own work and others I'm seeing features and velocity being chosen over everything else. The exhilaration of using these tools can be tremendous, leading to a ton of new features and code entering a codebase (myself included).
The downside of this is quality, complexity, and human comprehension. In my opinion there is a somewhat magical thinking belief that AI will save us from this. Yes, good harnesses formed with AI, AI agents helping with complexity, and traditional autonomy like tests and continual integration are very important. However, we are roughly 5 months into the post Claude 4.5 world that started in December when programming changed forever, and I think we personally over indexed into velocity at the expense of quality and comprehension and this mini-essay is me trying to lean back into those other directions, like a bit of a hangover after the sugar high of all this velocity.
Some argue that code doesn't matter anymore, that its a black box. Maybe this will be true in the future, but in this time interval I think code still matters (though I'm not writing by hand anymore, AI is). I think people who treat AI code as a black box are ahead of their skis -- this might be true in a year or two, but at this 6 month interval of time I don't believe it to be true. I think this is especially true in production code bases that involve multiple people working within it. In that case it's especially important to make sure there is a clear architecture and logic to the system, so that everyone can know how their work can fit into it.
There's a tendency to throw more AI at the problem ("just let AI handle the complexity!" "just let AI handle the quality!" etc.) While I do think it's important to layer AI onto these problems (you can indeed let AI help you manage the complexity and quality), we should also be careful to not have magical thinking about these AI systems, and that they will fix all of our problems ("why eat healthy and exercise when AI will just make me a magical drug that will solve all of my problems?").
So: now that we are 5 months into the Claude coding revolution, let's remember to think about quality, simplicity, and human comprehension again, not just velocity.
Agree to disagree, but this is my own personal viewpoint at this point in the Singularity. Yes, AI might magically save me from the flood of complexity and hidden issues I might have in my project (and I should be using AI as much as possible!), but its also true if I had just built my house on stilts with a good foundation then when the flood comes my house should be ok.
