Zero to AI Engineer in 2025: Compressing 10 Years of Industry Experience

In ten years, a lot has changed but plenty more has stayed the same in ML & AI.

To succeed you still need a mix of programming, maths, statistics and business skills. In this post and the accompanying Youtube video I wanted to share my two cents on how to get started in AI without breaking the bank.

Because with the amount of high quality of free resources available online, it's now easier than ever to get started in AI without spending USD 100k on an expensive degree, or USD 15k on an AI or data science bootcamp.

For more context, I recommend you watch the Youtube video first:

To be fair, I'm not a complete self-starter in AI – my graduate degree contained a lot of ML and statistics and it definitely helped me become an avid reader of ML papers.

Nevertheless, I did need to reinvent my profession a number of times over the past ten years because of all the changes in the field of data and AI. First with the rise of big data in 2014, and later when the deep learning revolution gained widespread industry adoption from 2017 onwards.

Today (in 2025) with AI agents all the hype, the same mix of skills remains as important as ever even though the context in which they are applied is of course quite different.

In what follows I tried to write down how I would learn AI engineering if I had to start from scratch. Taking advantage of resources we didn't have back when I started, and taking into account the rapid changes in AI we're seeing on an almost weekly basis. With the basis outlined in this blogpost, you should be able to navigate the world of AI for at least the next couple of years.

A fair warning. Even if most of the resources listed below are free or have a free version, if followed all the way through, this curriculum will take up quite a bit of time. Depending on your level of coding experience and how much time you have, anywhere from two to twelve months before you can work on the fun stuff. I do think that – regardless of how much no-code AI tools are made available – it still holds a competitive advantage to be able to code and read algorithms in 2025, even though the reasons why are different from 5 or 10 years ago.

Finally, I made sure that all the online components in this curriculum are free. A lot of the courses listed do have a paid track which allows you to get certifications. These can be a nice way to structure your learning, but certificates are generally not a requirement to get industry projects. A tangible track record of work, open source contributions or otherwise, is a much better indicator of what you're capable of than any digital certificate.

Lastly, I am aware that having your own laptop is not in reach for everyone. And even though most of the courses and online environments do offer browser-based programming environments, it might still be a good idea to see what your options are to get either a Linux or Macbook machine because a lot of AI tools require you to work from the command line and having control over your development environment can prevent a lot of headaches down the line.

All right, without further ado, let's get started!

Step 1: Code First, Ask Questions Later

First, fire up your AI-assisted IDE and start writing/generating messy Python code. Let the AI help you fix your code, and fix errors generated by your LLMs. Whether you're working in Cursor IDE, VSCode with Github Copilot, or Tabnine, it's like pair programming but your partner never gets tired! In 2025, you basically have a live-in AI coding instructor in your IDE, so why not use it!

... but you might ask, why Python? First of because it's by far the most widely used programming language in AI. Second, because you can write almost anything in Python - from AI algorithms, data processing logic, backend and frontend code! Finally, Python is also very beginner-friendly, especially compared to something like C++.

Some of the most common Python AI libraries used in industry are

• Pandas: a Python library for working with structured data ("dataframes").
• Scikit-learn: a swiss army knife of ML tools and algorithms for Python.
• Torch: one of the best Python libraries for deep learning algorithms.
• Transformers: the most extensive repository of OSS transformer models.
• Spacy: a very complete Python library for natural language processing.
• OpenCV: a popular Python library for computer vision.
• XGBoost: one of the best Python libraries for Gradient Boosting Trees.
• LangChain: one of the most popular AI agent frameworks.

Other Python libraries to check out:

• FastAPI: a great library for writing Python backend applications.
• Streamlit: a simple way to create frontends for your AI apps in Python.
• PySpark: to scale up your data processing with Apache Spark.

The greatest thing about Python is that all these libraries are free to use! They're Open Source Software (OSS, meaning you can use them both for private projects and commercial projects for free), community-maintained, and often best in class! What's not to like :)

Python programming starter kit

If you're new to programming, here are a couple of resources to get started:

• The Python programming language documentation.
• The Harvard edX Learning Python for Data Science course.
• Coursera's Programming for Everybody course.
• Linux Foundations' Introduction to Linux course.
• The Missing Semester of Your CS Education covers Git, Linux and Vim

As mentioned, I highly recommend using an AI-assisted IDE and the LLM of your choice as your AI tutor* - but that doesn't mean you should skimp on the basics of programming! Besides Python, it's probably also a good idea to know your way around a Linux terminal, since most AI tools and software is run and deployed in Linux environments. The Linux Foundation has a lot of good free resources, along with (paid) industry grade certifications.

*) As of this writing, the best LLM for coding is the latest release of Anthropic's Claude 3.5 Sonnet. Check LiveBench for updates.

Step 2: Get Your AI Basics Straight

Once you're comfortable writing Python code, it's time to dive into the foundations of AI. That means you'll need to get a good grasp of linear algebra, probability theory, statistics, and deep learning! It might sound like a lot, but with the right courses and instructors I'm confident you'll get there in no time!

Some good courses and resources to get started are

• AI Python for Beginners by Andrew Ng.
• MIT OpenCourseWare's Intro to ML course.
• The Coursera Mathematics for ML Specialization (esp. the Linear Algebra course)
• TU Delft's Probability Theory MOOC.
• Google Cloud's Machine Learning Engineering learning path.
• The Coursera Deep Learning Specialization (also by Andrew Ng).

Most of these presuppose some programming experience, so if you know how to code just jump right in-otherwise it's back to step one for you! All of the courses above are free to follow and audit for at least a couple of weeks. You can choose to get a paid course certificate, but that's definitely not a requirement to get a job in AI! (or start working on your own AI apps ;))

Bonus content - gems of ML literature

Some nice papers and free text books on AI and machine learning:

• R Fisher (1922). On the mathematical foundations of theoretical statistics.
• George EP Box (1976). Science and statistics.
• Sutton & Barto (2014). Reinforcement learning.
• Vaswani et al. (2017). Attention is all you need.
• Deisenroth et al (2020). Mathematics for machine learning.

... of course, this list is just skimming the surface - I'm sure you'll find your own way and ML domains once you get immersed!

Step 3: Join the AI Party

Once sitting in your room studying and implementing ML algorithms to lo-fi tunes is getting boring, it's time reach out into the world of ML! Luckily, there are plenty of awesome ML communities you can join for free to start building stuff, discuss the latest AI research and breakthroughs, or just share what you've been up to!

A small selection from the ever-growing list of ML communities:

• The HuggingFace forums: where OSS AI enthusiasts and contributors discuss training and fine-tuning LLMs and other language models.
• EleutherAI's Discord chat rooms: another vibrant community of open sources AI enthusiasts, researchers and contributors.
• Another great place to find AI folks is on X. Check out Robert Scoble's amazing lists to find the top AI researchers on X.
• You could also decide to attend a top-tier ML conference like ICLR or ICML. You'll find that most AI academics are very open to new voices!

... and of course, every one of the open source Python libraries mentioned above has its own community of contributors, coaches and users. Helping out the AI OSS community by contributing to widely used Python AI libraries is a great way to make friends!

If you haven't, this is also a great opportunity to start playing around with AI assistants. Over the course of 2024, I've had paid licenses for the following AI tools:

• OpenAI's ChatGPT. Ended up cancelling because GPT-4o is not quite as good as Claude 3.5 Sonnet.
• Anthropic's Claude AI. Ended up cancelling because it doesn't do internet searches.
• Perplexity AI. This is my current go-to AI assistant, I use it with Claude 3.5 Sonnet.
• Cursor IDE. This IDE dramatically changed how I develop ever since downloading it back in June 2024, and I'm still a daily user.

Right now (December 2024), I'm mostly using Claude 3.5 Sonnet for both AI coding and content generation, but that will probably change in 2025! One way - besides the AI community - to keep up with developments in AI is through benchmarks like

• The LMSys Chatbot Arena: it benchmarks LLM-based AI assistants on the quality of their responses by head-to-head comparisons from human evaluators.
• SimpleBench: a benchmark of LLM reasoning capabilities - which is key to developing high-quality trustworthy AI agents!
• LiveBench: a continuously updated benchmark of LLM capabilities.
• For traditional ML tasks I often check ML model and method benchmarks on Papers With Code, although for most tasks these days the SoTA (State of The Art) is an LLM.

Step 4: Get Your Hands Dirty

But in AI nothing beats actually building stuff. Most tooling is free, and you have your pick of free environments to work in. The hardest part finding something interesting to build! Maybe you want to build an AI model that helps with your content creation, or an AI tool that makes your life just a little bit more fun. The key is to build something you actually care about. Bonus points if it also create value (for example a consumer AI app, or if it solves a business problem) or pushes the envelope when it comes to the state of the art in AI (for example by fine-tuning an LLM). The more value you create with the problems you solve, the more your experience will be worth!

If you own a laptop, IDEs like Cursor or VSCode are a great place to start. If you're thinking of getting a laptop for AI stuff, right nowMac books are still your best bet. But if you don't have the money to buy your own laptop, there are some great free online AI environments in which you can get started right away!

• Google Colab offers free AI notebooks.
• Kaggle offers free resources along with an active AI community.

That being said, for more comprehensive AI projects having your own laptop will be a huge benefit. At some points, running all your code in notebook environments becomes unwieldy and hard to maintain, and you might start thinking about combining code and packaging AI Python projects using tools like Docker and FastAPI.

Some of the tooling I always have running in my own development environments:

• An IDE (Integrated Development Environment). Right now I'm mostly using Cursor, which is an IDE with AI capabilities baked in. I've also used PyCharm and VSCode over the years. This is a fast moving field, so expect things to change in 2025.
• Terminal access for command line tools (iTerm for Mac, or via your IDE). See the The Linux Foundation course mentioned in step 1 for a good introduction to the command line.
• Docker. This is the industry standard way to run and deploy AI or other applications in encapsulated runtimes.
• Git. Version control for your code repositories / AI projects.

And that's it! You will need to install programming languages like Python and create environments to manage dependencies for different projects you're working on, but mastering those four tools - all of which are free or have free versions available to download - will get you a long way towards becoming a professional AI engineer.

Step 5: Stay in the Loop

Finally, developments in AI are moving at a breakneck speeds. Pretty much every week a new AI tool, model or development is released! My recommendation here is to use newsletters to read up on AI news once a week. Getting your AI news daily is probably overkill for the speed at which AI developments and innovations are made. Plus you probably also have more important stuff to do - like build your own fantastic AI apps :)

Some great AI newsletters include:

• AlphaSignal (weekly, technical AI news)
• Ben's Bites (weekly, AI tools and AI news)
• The Neuron (daily, AI advancements)
• The Rundown AI (daily, AI business & AI news)

Alternatively, X is a great place to keep up with AI news since a lot of the top AI researchers are active on there. Robert Scoble created a bunch of lists with the who's who and what's what for the AI community on X.

Remember: You don't need to understand everything. Even the experts are asking their AI assistants for stuff daily. Just start building, keep learning, and have fun with it. The best AI engineers I know are the ones who treat it like a playground rather than a textbook.

P.S. If anyone tells you that you need a $20K bootcamp to learn AI, send them this post and save them some money. 😉