Should I learn AI and machine learning, or become a strong software engineer first?

First, separate the two careers

When students say "I want to go into AI", they usually mean one of two quite different jobs.

Most of the jobs that have appeared in the last two years are in the left column. And the left column is, overwhelmingly, software engineering with a model in the middle.

Why I usually say engineering first

When a company builds an AI feature, the model call is perhaps ten percent of the work. The rest is ordinary engineering: getting the right data to the model, handling failures and timeouts, storing results, checking quality, controlling cost, protecting user data, and deploying it reliably. A person who is weak at those things cannot ship an AI product, no matter how many machine learning courses they have finished.

I have seen many students with three ML certificates who cannot build a working API or deploy anything. They are not hireable for either column. I rarely see the opposite problem, because a good engineer can learn to use models in weeks.

When going straight to machine learning makes sense

There are people for whom ML first is the right call. You probably fit if most of these are true:

1. You genuinely enjoy maths and statistics, not just the idea of AI.
2. You are willing to do a master's or research-focused path.
3. You like reading papers and running experiments more than building products.
4. You are comfortable with a smaller, more competitive job market in exchange for deeper work.

If that describes you, lean in, but still learn to write clean, tested code. Research teams complain constantly about code that nobody else can run.

A path that does not force you to choose

For most students I suggest this order:

1. Months 1 to 4: get solid at one language and one stack. Build and deploy two complete projects with a database, authentication and a clean README.
2. Months 4 to 6: learn the fundamentals that sit under everything: HTTP, databases, basic system design, Git, testing.
3. Months 6 to 8: add AI as a layer. Use an LLM API in one of your projects. Add retrieval over your own data. Measure whether the output is actually good.
4. After that: decide based on what you enjoyed. If you loved the evaluation and data side, go deeper into ML. If you loved shipping the product, you are an AI-aware software engineer, which is exactly what most companies want.

How to know you are ready to add AI

You do not need to be an expert engineer. A good signal is when you can build a small application end to end without following a tutorial, you understand what an API call does and how to handle it failing, and you can read an error message and make progress on your own. From that point, AI topics stop being magic and become tools you can reason about.

The fear of being left behind

The worry underneath this question is usually "if I spend a year on fundamentals, everyone else will be ahead". In practice, the people rushing into AI without foundations stall quickly, and the ones with foundations catch up in a few weeks. A year of real engineering is not a delay; it is the thing that makes every later topic faster.

A self-check before you decide

Answer these honestly, yes or no:

1. Could I build a small web application with a database and deploy it without a tutorial?
2. Have I enjoyed a maths or statistics course for its own sake, not just passed it?
3. When I imagine my ideal week, is it mostly running experiments, or mostly shipping features people use?
4. Would I be willing to spend two more years in formal study to reach the role I want?

If you said no to the first question, engineering first is almost certainly right, whatever you answer to the rest. If you said yes to the first and yes to two and four, a research-leaning ML path is worth exploring seriously. If you said yes to the first and the third answer was "shipping features", you are describing an AI engineer, and your next step is adding AI to real products.

An example of the combined route

Imagine a student who spends six months building a clinic appointment system: patients book slots, staff manage schedules, reminders go out by email. Nothing about it says AI. Then they add one feature: patients describe their symptoms in free text, and the system suggests which department to book, with a clear disclaimer and a human able to override. They build a test set of fifty sample messages and measure accuracy.

That student can now talk credibly to a product company, an AI startup, or a health technology team. They did not have to choose between the two columns. They built the engineering base and then showed they could apply a model responsibly.

If you are still stuck

Read An AI Engineering Roadmap That Puts Things in the Right Order and The AI Engineering Words You Keep Hearing, Explained Properly. If you want help deciding which column fits you, join Sefism and, as a member, book a 1:1 session.