Maria had two offers on the table. One was a $62,000 master’s in artificial intelligence at a private university with a leafy campus and a very glossy brochure. The other was an online program from a large state school that would cost her under $11,000 in total. Same three words on the diploma. What separated them wasn’t prestige. It was that she spent three weeks reading syllabi, emailing professors, and asking a deeply unglamorous question: which of these two puts me in a room with a hiring manager six months after graduation?
She picked the cheap one. Fourteen months later she was a machine learning engineer at a healthcare startup, working on computer vision for pathology slides.
Choosing among AI degree programs is mostly a research project you run on yourself. Here’s the process, step by step, with the numbers and questions that actually move the decision.
Step 1: Name the job you’re training for
“AI” describes a job about as precisely as “science” does. Four roles dominate the hiring market, and they want different coursework.
- Machine learning engineer. Ships models to production. Needs containers, CI/CD, and model monitoring as much as modeling.
- Data scientist. Analytics, SQL, experiment design, causal inference. Often closer to statistics than to deep learning.
- Applied scientist. Industry research labs. Publishes, prototypes, then hands the model to engineers.
- Research scientist. Almost always a PhD. Conference papers are the real currency.
Two of those four overlap by maybe 60% in coursework. The electives diverge fast. Write your target role at the top of a page before you open a single program website, because every later comparison gets measured against it.
Step 2: Read the course list, not the marketing page
Find the degree requirements page and count how many required courses are actually technical. A program with two intro AI courses and eight business electives is a data analytics degree wearing a trendier label.
Look for titles like Deep Learning, Natural Language Processing, Reinforcement Learning, Probabilistic Graphical Models, and Computer Vision, plus at least one systems course covering distributed computing or model deployment. Those names tell you more than any accreditation badge.
Then pull three syllabi as PDFs and check the assignments. Do students train models on real datasets, or write essays about algorithmic bias? Both are legitimate degrees. Only one gets you through a technical interview. If you want a wider benchmark on curricula, tuition, and outcomes across the field, this breakdown of what AI degree programs really teach and what they cost is worth reading before you shortlist anything.
Step 3: Audit your prerequisites honestly
Most graduate AI programs quietly assume you arrive with:
- Calculus I through III, including multivariable
- Linear algebra
- Probability and statistics
- Data structures and algorithms
- Fluency in Python, plus one compiled language
Missing linear algebra is the most common gap, and it’s the one that breaks people. A graduate ML course hands you matrix notation in week one and never slows down. Fix it cheaply: a community college course runs a few hundred dollars, and a post-bacc enrollment gets you the transcript line. Do that in the six months before applications open, not the month after you’re admitted.
Step 4: Run the money math properly
Sticker price is the least useful number. Compare net cost instead.
Georgia Tech’s online master’s charges roughly $180 per credit hour, landing the whole degree near $7,000 before fees. UT Austin’s online computer science master’s sits around $10,000. A private on-campus program can run $60,000 to $80,000. State flagships swing hard on residency, often $16,000 in-state against $55,000 or more out-of-state each year.
Then subtract. Research and teaching assistantships frequently waive tuition and add a stipend, especially in PhD tracks. Many US employers reimburse up to $5,250 a year tax-free.
The cost line nobody writes down
Forgone salary. Two years out of a $95,000 job costs $190,000 in wages you never earned, and that line usually dwarfs tuition. Put it in the spreadsheet next to the tuition figure, because it changes which programs look affordable.
Step 5: Test-drive the material before you commit
Take one hard course from the exact curriculum you’re considering, for free, before you pay a deposit.
Test-drive for under $50
If computer vision sits on your list, work through a short project-based sequence first. These OpenCV AI courses are a decent proxy for whether you enjoy the actual work or just the idea of it. Twelve weeks of weekly problem sets exposes a fake interest fast.
Studying has also gotten cheaper. Using Meta’s AI as a study partner for derivations, debugging, and paper summaries can compress a week of confusion into an afternoon, provided you still do the problems by hand before the exam.
Step 6: Interview two recent graduates
Skip admissions. Search LinkedIn for the program name plus “machine learning engineer” and filter for people who graduated in the last 18 months. Ask three questions:
- How many hours a week did the core courses actually take?
- Did your capstone become something you showed in interviews?
- What do the classmates who didn’t land AI roles have in common?
That last one is the money question. Nobody answers it the same way twice.
Step 7: Verify where graduates actually land
Ask for the placement report, not the “career outcomes” web page. You want graduate count, response rate, named employers, and median base salary. A school advertising “94% employed within six months” without disclosing how many people graduated is answering a different question than the one you asked.
For research-heavy options, look at lab output and industry partnerships, since those determine which projects you can actually join. Columbia Engineering’s AI programs are a useful case study in how faculty research groups, location, and corporate recruiting reinforce each other.
When a full degree isn’t the right call yet
Sometimes the fastest route into AI work isn’t a two-year degree. Product-focused builders often get further with a short, intense program than with a thesis. Places like Cerebral Valley Academy exist specifically for people who want to ship working software in weeks rather than semesters. That doesn’t replace a master’s if you’re aiming at research. It can get you hired sooner if you’re aiming at implementation.
Your 90-day plan before applications close
Fall cohort deadlines usually land between December and February. Work backwards.
Days 1 to 30: enroll in the missing prerequisite. Start one free module from your target curriculum and finish it, not half of it.
Days 31 to 60: draft your statement of purpose around one project instead of one ambition. Rebuild a small model end to end, data cleaning through evaluation, and write the results down with numbers.
Days 61 to 90: request transcripts, line up two recommenders who can speak to technical work, and message three current students. Submit two weeks early, because recommenders are humans with calendars.
Then protect the first semester once you’re in. The students who struggle aren’t the ones short on talent. They’re the ones who took four heavy courses while working full time and stopped sleeping in week six. Take two. Get an A. Build one thing you can show someone.