How I'd Become an AI Engineer in 2026 (Even with No CS Degree) Transcript, AI Summary & Key Points

00:00 This is how I'd become an AI engineer in 2026 if I was starting from zero today. I'm a senior applied scientist at Twitch where I build production AI systems, and my degrees are in politics. So, I got into this a weird way myself, and I've coached hundreds of people from every kind of background into AI and ML roles. So, I know that it's possible, but only with the right strategy.

00:17 In this video, we'll go through what an AI engineer actually does and why the salaries are so high. Then we'll go over the skills you need, how to learn them without getting stuck in tutorial hell, what to build for projects, and how to get a job in the field even with resume scanners working against you. Let's start with what this role does day-to-day because the titles can be a bit unclear.

00:35 AI engineer, machine learning engineer, and software engineer all kind of blur together in job postings. To put it simply, an AI engineer builds applications on top of models that somebody else already trained. So, think of Claude, GPT, Gemini, or open-source models. They use these models to create systems that summarize information or automate tasks.

00:54 This is different from a machine learning engineer who uses data to train custom models, like a recommendation system for Netflix or a fraud detection model. Neither ML engineers or AI engineers invent new kinds of models. That's work for AI researchers. Those guys have PhDs in a totally different career track, so we won't worry about them in this video.

01:14 Software engineers build the application around the model, so the interface, API, or back-end. They don't own anything specific to the models at all. A lot of people think that AI engineers are just software engineers calling a different API. To that I say, that is only true if you are a bad AI engineer. There's a lot more going on besides the API call, which is what this video is about.

01:33 That gap between just calling an API and building robust AI systems is a big part of why AI engineer salaries are so high. There are a lot of companies wanting to integrate AI into their products right now, and surprisingly few people who can do that job well. The good news is you do not need to master every single thing in AI before you can start building, but you do need to know what matters because this field is huge.

01:55 And if you just start learning whatever Reddit or YouTube tells you is important that week, you can spend a ridiculous amount of time on stuff that barely matters for the job you actually want. So, I think about the AI engineer skill set in seven buckets. [music] This is a pretty long list, so I put it all in a PDF you can download for free in the description.

02:12 The first skill bucket is programming. You need to get really good at Python. This is the language you'll be using to code daily, so you should be comfortable with the fundamentals like functions, classes, data structures, error handling, working with APIs, and using third-party libraries. While you learn Python, pay attention to core software engineering skills, too.

02:31 Like structuring a project, using Git, working with virtual environments and config files, writing tests, and navigating from the command line. AI engineers need solid software engineering fundamentals, but you do not need to be an expert in back-end engineering. For basically all of my personal projects, I use Supabase for the back-end. It gives me a Postgres database, authentication, APIs, storage, and a bunch of other stuff without me having to do all of that from scratch.

02:57 But this actually creates an interesting problem when you're coding with AI. Supabase changes all the time. There are current CLI commands, security requirements, and best practices that your coding agent might not know just from its training data, which is why Supabase made agent skills. A skill is basically a set of instructions written in plain markdown that teaches your coding agent how to do a specific job properly.

03:18 When you're doing something relevant, the agent can load those instructions into its context and use them while it works. So, Supabase skills include things like their current security requirements and CLI commands, but importantly, they also tell the agent to check the live Supabase docs before writing code instead of assuming whatever it learned during training is still correct.

03:36 You can install them with one command, and they work with coding agents like Claude Code, Cursor, and others. And you'd be surprised how much of a difference it makes. Like here, I asked Claude Code without the skill to review my schema for performance issues, and it said there aren't any. Hooray! Then I tried again with the Superbase agent skills and it turned out Claude Code had just missed a lot of really important problems.

03:56 Learning how to code with AI the right way [music] is really important. An effective use of skills like this makes a big difference. The link is in the description if you want to try it out. Thanks to Superbase for sponsoring this video. Now that we understand the software engineering requirements, let's talk about the second skill, math. And the good news is you need way less math for AI engineering than you probably think.

04:15 It will be helpful for you to understand the basics of linear algebra, probability, statistics, and calculus. But for this role, I'd focus way more on building intuition than doing complicated math by hand. You want to understand what these concepts mean and recognize when they show up in the systems you're working with, but you will never need to calculate things by hand, I promise you.

04:36 >> [music] >> The third skill is machine learning fundamentals. Now remember I said this role is not about training models from scratch. That's completely true, but some core concepts will probably come up even when you're using pre-trained models. So things like supervised versus unsupervised learning, training and test sets, overfitting and underfitting, and common evaluation metrics like precision and recall are good to get under your belt.

04:57 You should also understand neural networks at a high level and eventually how transformers work, since they're the architecture behind most of the models you'll be working [music] with. Again, the goal here isn't to become a machine learning researcher. You just want enough understanding that you can kind of differentiate between models and pick the right one for your use case.

05:11 The fourth skill bucket is where we get into actual AI engineering fundamentals. First, you need to get comfortable working with LLMs. That means using the major model options, understanding the different models available, and knowing how to choose between them based on things like performance, cost, speed, and licensing. You'll also need prompt engineering, and I don't just mean messing around with a prompt until the answer looks good.

05:32 You should understand things like few-shot prompting, structured outputs, defensive prompting, and how to systematically test whether your changes are actually making things better. Then there's context engineering, which has become a really important part of building AI systems. This is basically deciding what information the model gets to see and what gets left out when you have too much.

05:50 You'll also want to understand rag or retrieval augmented generation. This is when we allow our model to see data that wasn't in its training set like internal company docs. Getting good at this means understanding things like embeddings, chunking, vector databases, and how to retrieve the right information to give your model. And finally, evaluation.

06:07 You need a way to actually measure whether your AI system is doing what you want it to do using things like test sets, appropriate metrics, LLM as judge, and human evaluation. These last few skills are where we start getting beyond just calling an API and actually engineering the system around the model. And we'll come back to all of them later when we take a simple project and turn it into something you can actually put in front of users.

06:28 Now we're onto the fifth skill bucket, which is more advanced AI engineering applications. Specifically, AI agents. An agent is basically an LLM that can call functions and keep taking action until it completes a task. So instead of just generating text, it might search the web, query a database, call an API, or write and execute code. You'll want to understand how tool calling works, how to give an agent access to external systems, and how to evaluate whether it's actually completing tasks correctly.

06:52 You should also learn MCP or model context protocol, which is a standard for connecting AI applications to external tools and data. And eventually, you'll want to get into multi-agent systems where multiple agents work together on different parts of a task. This is one of the areas that's changed the most over the last couple of years and it's becoming a pretty important part of what AI engineers are expected to know.

07:13 Now we're onto the sixth skill bucket, [music] production engineering and infrastructure. Building an app that works on your laptop is one thing. Building one that can reliably handle real users is a whole different set of skills. For this, you'll want to understand how to build APIs, use cloud platforms, containerize your applications with Docker, and set up basic CI/CD so you can actually deploy changes confidently.

07:33 You'll also need monitoring and logging so you know what your system is doing once it's running and can figure out what went wrong when something breaks. This is one of the most important skill buckets to spend time on because it's what takes you from being able to build just like a little demo to something that people can actually use. And final skill bucket is interview prep.

07:50 Unfortunately, knowing how to do the job and knowing how to pass the interview are still two different skills. For coding interviews, you'll want to practice data structures and algorithms and get comfortable solving leak code style problems in Python. How much of this you need varies a lot by company, but it's still pretty common in technical interviews.

08:07 You'll also want to prepare for system design interviews. And for AI engineering roles specifically, that means gen AI system design. So instead of just being asked to design something like Twitter or Uber, you might be asked how you design a rag system, a customer support agent, or some other LLM application that has to work at scale. I would work on interview prep in parallel with everything else, especially leak code, rather than waiting until you feel completely ready to apply.

08:31 Otherwise, you can spend months getting better at AI engineering and then realize you're totally unprepared for the actual interview. Now looking at this list, you might be thinking, "Okay, that is a lot of stuff to learn." And it is. That's why these jobs pay so well. But the biggest mistake you can make here is thinking you need to sit down and learn all seven of those buckets before you're ready to build anything.

08:50 I actually recommend almost the opposite. Start with a course, book, or video to get the basics and understand what's out there. Then start building something as quickly as you can. As you build, you're going to run into things you don't understand. That's when you go back and learn those things in more depth. We'll talk more about projects in the next section, but here are some resources to get the lay of the land before you dive in.

09:10 I'm mostly going to focus on things that are free or pretty inexpensive because you absolutely do not need to spend thousands of dollars learning this stuff. For Python, I'd start with Scrimba's learn Python course. [music] It's free to get started, super interactive, and it'll actually get you writing Python instead of just sitting there watching someone else code.

09:26 For math, 3Blue1Brown, no question at all. The Essence of Linear Algebra and Essence of Calculus series are both free on YouTube, and they're really good at giving you that visual intuition we talked about without making you spend months doing math by hand. For machine learning fundamentals, I'd use Google's machine learning crash course. It's free, [music] fairly short, and covers most of the core concepts you actually need to get started.

09:47 Then once you get into the actual AI engineering stuff, there are a lot of resources. Deeplearning.AI has a bunch of short courses on things like prompting, rag, agents, and MCP, and many of them are free to take. I like these because you can pick them up as you encounter that specific topic in a project instead of committing to another giant program.

10:06 These are also free to audit. And Hugging Face has a free LLM course if you want to understand language models in more depth. Finally, if you're going to buy one book, I'd make it AI engineering by Chip Huyen. It covers a huge portion of what we just talked about, including working with foundation models, evaluation, rag, and actually building these systems for production.

10:24 The one caveat is the AI engineering came out before a lot of the recent work on MCP and agents because this field moves ridiculously fast. So if you're feeling ambitious and want something more up-to-date and more challenging, check out the free Hitchhiker's Guide to AI Engineering. I'll put the link to that in the description as well. And again, you do not need to finish all of these before you move on.

10:41 Pick the resource that covers what you need right now. Learn enough to get moving and then go build something, which brings us to your first project. So for your first project, I want you to build something really simple. And I mean really simple. Your first project does not need to impress a hiring manager. It does not need thousands of users or some crazy agent architecture or an idea nobody has ever thought about before.

11:02 It just needs to help you get started. The point of this project is to take all of the concepts you've been learning and actually use them. Because this is where you're going to find out what you actually understand and what you might need to go back and review. So I'd pick a problem that's personally interesting to you and small enough that you can get a first version working really quickly.

11:21 Maybe you can build a recipe bot that connects to scans of your family recipes and suggest dinner options. Or an automation to read the news and send you a daily brief. The The idea really doesn't matter that much. What matters is that there's some personal connection to it. When you already understand the subject, you have a much better sense of whether your AI system is actually doing a good job, and you're way more likely to keep working on it once you get past that first fun couple of hours and into the annoying

11:45 debugging part. And I do think it's important to struggle a bit with this first project. It's okay to use AI to explain concepts that are new, but I really recommend coding by hand and trying to figure it out yourself as you go. I've seen as a manager and on hiring teams that folks who join the job market post-AI are really limiting themselves by using AI before they're ready, and then they never reach their own potential.

12:05 By the way, if you want more advice on leveling up in AI without the hype, make sure to subscribe. [music] So, once you have this simple project down, it's time for something more advanced and more [music] fun. This is where I'd go back and review that skills list we talked about earlier and start adding anything you missed to your project. Build proper evals so you can measure whether it's actually working.

12:24 Experiment with context engineering and rag. Give your model tools and turn it into an agent if the problem makes sense for that. Then, take it all the way to production. Deploy it somewhere people can use, add monitoring and logging, and start thinking about things like cost, speed, security, and what happens when real users start behaving in ways you weren't expecting.

12:41 You don't need to cram every AI buzzword into one project. The goal is to show that you can take an AI system from an idea to something reliable that people can use. And if you want help doing exactly that, stick around to the end because I have a really cool announcement. At this point, you've proven you can build something and you have the skills to do the job.

13:00 But companies aren't really hiring you because you have a list of skills. They're hiring you because they have problems and they need your help. With your project, you've proven you can solve your own problem. So, [music] the next step is to go solve someone else's. And you don't need to wait for somebody to hire you to do that. You could volunteer for a nonprofit and build something that saves them time.

13:16 Build a tool for a hobby group you're already part of. Help a friend's small business automate some annoying process. Or contribute to an open source project where you're working on legit problems with people who are more experienced and can give you feedback. The specific route doesn't matter so much. What you're trying to get is experience working with requirements you didn't make up, constraints you didn't choose, and ideally people who are actually going to use it.

13:38 Because now when a hiring manager asks about your experience, you have something much stronger to talk about. And if you do it right, it can even get listed as experience on your resume. Now to be clear, I am not telling you to make an AI startup just to get a job. I am just encouraging you to be creative about finding ways to make your own experience without waiting for someone to take a chance on you.

13:58 Because the reality is that breaking into this field without a traditional background is hard. You're going to need to get creative, be brave, and put yourself out there. Which takes us to the final step, getting a job when your resume won't pass the automated screener. So if you don't have a relevant degree or traditional engineering experience, the reality is that your resume is probably going to get filtered out of some jobs before a human ever sees it.

14:19 [music] That sucks, but it doesn't mean you can't get hired. It just means the traditional method of applying on LinkedIn probably isn't going to work. You'll need to get creative with the networking. And when I say networking, I don't mean going to some awkward giant conference and trying to convince strangers to give you a job. I mean actually getting to know people who work at the companies you're interested in.

14:38 Pick a handful of companies you'd really want to work for. Find AI engineers, hiring managers, or other technical people there who post online, contribute to open source, give talks, or have some kind of public work that you can learn from. Then learn about that work and engage with them like a normal person. Ask a question about something they built.

14:54 Share something useful related to their work. If you've built something relevant, show them. The goal is not to immediately ask for a referral. It's to build enough of a real relationship that when a role eventually opens up, you're not just some random resume in a pile of thousands. I got my first internship by cold emailing a company in a country where I didn't speak the language while I was studying something completely unrelated.

15:16 So I know this can work. But I also know that doing all of this by yourself is hard, especially that jump we talked about earlier where you have the fundamentals down, but you're trying to go from building simple projects to something that actually demonstrates you can do AI engineering professionally, which is why I'm putting together an 8-week cohort where I'm going to personally work with a super small group of people to build their own production-ready AI engineering projects.

15:40 This is not for complete beginners. You should already be comfortable with Python and have some of the AI fundamentals we talked about today. But if you're at that point where you know the basics and you're not sure how to turn them into something that actually gets you closer to a job, that's exactly what we're going to work on. We'll go through choosing and scoping a project, building it, and taking it all the way through deployment.

15:59 And because this is the first time I'm doing this, I'm keeping the group really small. It'll be application only, and this will probably be the most hands-on one-on-one time I have ever offered or maybe ever will. So if that sounds like where you are right now, I'll put the application in the description. [music] And if you're ready to learn more about the different kinds of AI engineering projects, check out my AI engineering project roadmap at my