2026 6 step roadmap

This is your roadmap to engineer in this new AI world and buckle up, this video may be a little bit longer than usual, but these are the six steps that I think are necessary to learn. Stop learning Python in notebooks. Step 1: Download Claude Code, Cursor, Codex and build something. Actually give it a project here. It's not to code, it's actually going to be to supervise. Oh my god, saturated. Understand every single line or most of the code that's being generated. Step 2: Build stuff and throw it away. It's counterintuitive, but build a bunch of throwaway projects that you're gonna end up scraping. But the point of this part is to actually get reps. It can be a prediction model, some RAG app, some website app, doesn't have to be necessarily useful, but just make it functional and actually get your reps in. Step 3: Test your coding capabilities. YouTube, zero to hero and actually hand code by the neural networks back propagation. I can't believe I have to say hand code when this is how I learned how to code lol. By the neural networks because that's where a lot of the learning of the fundamentals is actually lived. The point is to hand code it and then debug it so that you understand what's going wrong and how to fix it. Step 4: Learn Evals. Probably the rarest skill and I think the most important. Luckily, I was able to learn this in my corporate job and that's Evals. Models will actually fail silently in production benchmarks are really good when they're in demo and then stuff happens and then they end up a fail or a model will tell you that it's 99% accurate, but the data has leakage or the data is not all clean. Understanding how to evaluate models is very critical. Step 5: Pick a niche/lane you're interested in and build a REAL WORLD PROBLEM. Sorry for caps, it's just important. It could be that you're interested in LLM engineering, be it Evals or it could be ML production. Kind of pick a lane and build one really big serious project around it. That's what you can put on your resume. The requirement is that it should be live, monitored, Evals documented. Well, last step is Step 6: Create content around it and share it with the world. You must write everything down. All documented. What you liked about it, what didn't like about it, what broke and how you fixed it because right now judgment is really the uniqueness that you have in this day and age and being able to distribute what you've learned is really, really beneficial and helpful to other people as well.