How I'd Become an AI Engineer in 2026 (Even with No CS Degree)

Marina Wyss is a senior applied scientist at Twitch who builds production AI systems, and her degrees are in politics. Here's how she'd become an AI engineer starting from zero today:

1. Know the role. An AI engineer builds applications on top of models someone else trained (Claude, GPT, Gemini, open-source models).
2. Learn 7 skill buckets: Python, math, ML basics, AI engineering, agents, production and interview prep. Learn each only well enough to start.
3. Build before you feel ready. Get the basics, start a project, and go back to learn whatever gets you stuck.
4. Start with a simple personal project, then take it to production with evals, RAG or tools, deployment and monitoring.
5. Solve someone else's real problem to get experience with real requirements and users.
6. Network around the resume screeners by building real relationships with engineers at a few target companies.

Job titles blur together in postings. Here's how she separates them:

• Lots of companies want to integrate AI into their products, and surprisingly few people can do it well.
• You don't need to master everything in AI, but you need to know what matters. Learning whatever Reddit or YouTube hypes that week wastes huge amounts of time.

1. Python and software engineering

• Get really good at Python: functions, classes, data structures, error handling, working with APIs and third-party libraries.
• Learn the engineering basics alongside it: project structure, Git, virtual environments, config files, tests and the command line.
• You don't need to be a back-end expert. She uses Supabase (Postgres, auth, APIs, storage) for her personal projects.
• Tip for coding with AI: tools change faster than a model's training data. Give your coding agent skills (plain markdown instructions) that make it check the live docs. Without one, Claude Code told her a schema had no performance issues; with the Supabase skill, it found several.

2. Math

Far less than you think. Understand the basics of linear algebra, probability, statistics and calculus as intuition. "You will never need to calculate things by hand, I promise you."

3. Machine learning fundamentals

You won't train models from scratch, but these come up anyway: supervised vs unsupervised learning, training and test sets, overfitting and underfitting, precision and recall, plus neural networks and transformers at a high level. The goal is to tell models apart and pick the right one.

4. AI engineering fundamentals

• LLMs: know the major models and choose on performance, cost, speed and licensing.
• Prompt engineering, done properly: few-shot prompting, structured outputs, defensive prompting, and systematically testing whether changes help.
• Context engineering: deciding what the model sees and what gets left out when there's too much.
• RAG (retrieval-augmented generation): letting the model use data outside its training set, like internal docs. Covers embeddings, chunking, vector databases and retrieval.
• Evaluation: test sets, the right metrics, LLM-as-judge and human evaluation.

5. AI agents

An agent is an LLM that can call functions and keep acting until a task is done: search the web, query a database, call an API, run code. Learn tool calling, how to evaluate task completion, MCP (Model Context Protocol) for connecting to external tools and data, and eventually multi-agent systems. This area has changed the most recently.

6. Production engineering and infrastructure

Working on your laptop isn't the same as handling real users. Learn APIs, cloud platforms, Docker, basic CI/CD, monitoring and logging. She calls this one of the most important buckets, because it's what turns a demo into something people can use.

7. Interview prep

• Doing the job and passing the interview are different skills.
• Coding: data structures and algorithms, LeetCode-style problems in Python.
• Gen AI system design: instead of "design Twitter," expect "design a RAG system" or "design a customer support agent" that works at scale.

You don't need to spend thousands of dollars. Her picks, mostly free:

Pick the resource that covers what you need right now, learn enough to get moving, then go build.

• It doesn't need to impress a hiring manager, have thousands of users, use a fancy agent architecture or be an original idea. It just needs to get you started.
• Its job is to show you what you actually understand and what you need to review.
• Pick something personal and small enough to get a first version working quickly, like a bot that suggests dinners from scans of your family recipes, or a daily news brief.
• A personal connection matters. You can judge whether the AI is doing a good job, and you'll keep going once the fun first hours turn into debugging.

Go back to the skills list and add whatever you missed:

• Build proper evals to measure whether it actually works.
• Experiment with context engineering and RAG.
• Give the model tools and make it an agent, if the problem calls for it.
• Deploy it where people can use it, with monitoring and logging.
• Think about cost, speed, security, and what happens when real users behave in ways you didn't expect.

Don't cram every buzzword into one project. The goal is to show you can take an AI system from an idea to something reliable that people use.

Companies don't hire a list of skills; they hire people to solve their problems. Your project solved your own problem, so the next step is solving someone else's, without waiting to be hired:

• Build something that saves time for a nonprofit.
• Build a tool for a hobby group you're already in.
• Automate an annoying process for a friend's small business.
• Contribute to open source, working on real problems with experienced people who can give feedback.

The route doesn't matter much. What counts is working with requirements you didn't make up, constraints you didn't choose and people who actually use it. Done right, it can go on your resume as experience. She's not telling you to start an AI startup; just be creative about making your own experience.

• Without a relevant degree or engineering experience, your resume will often be filtered out before a human sees it. That means mass-applying on LinkedIn probably won't work.
• Networking doesn't mean awkward giant conferences. It means getting to know people at companies you want to work for.
• Pick a handful of companies. Find AI engineers, hiring managers or technical people there who post online, contribute to open source, give talks or have other public work.
• Engage like a normal person: ask about something they built, share something useful related to their work, or show them something relevant you've built.
• Don't ask for a referral straight away. Build a real enough relationship that when a role opens, you're not a random resume in a pile of thousands.

She's running a small, application-only 8-week cohort to help people make the jump from simple projects to work that shows professional AI engineering. It's not for complete beginners: you should already be comfortable with Python and some AI fundamentals. It covers choosing and scoping a project, building it and deploying it. The application link is in the video description.