Math Academy review for adults: worth it if you want to do math
I have not completed a Math Academy course. This review comes from Math Academy's own pages, from Justin Skycak's free book The Math Academy Way, and from adults who used it and put a name or a handle on what they wrote. I build Understand, a voice tutor that works in a very different way, so weigh my conclusions accordingly.
In my comparison of AI tutors I told anyone who wants to be able to do math to use Math Academy over my own product. I still think that. My disagreement is about the order in which an adult should learn things.
Math Academy in brief
Math Academy is an online math course with no videos and no human teacher. You read a short worked example, solve problems, and the software decides what you see next. It runs from fourth-grade math to university courses such as Linear Algebra, Multivariable Calculus and Methods of Proof.
The founders are Jason and Sandy Roberts. Jason studied mathematics at the University of Chicago and, by the company's account, built much of Uber's original real-time technology as a consultant after turning down the CTO job. The two started by coaching a fourth-grade math team in Pasadena and were teaching the same children calculus by the end of fifth grade. That grew into a program inside the Pasadena school district, and the homework software Jason wrote for it became the product when the pandemic moved everything online. Justin Skycak is Chief Quant and Director of Analytics. By his own account he built the system that chooses each student's tasks. He also wrote the book, a 510-page working draft last updated on August 13, 2026.
It costs $49 a month per student. The FAQ says there is no free trial and no regional discount, a full refund in the first 30 days, and a pause of up to 90 days. The site still says "Join Beta".
The diagnostic and the knowledge graph
You pick a course and take an adaptive diagnostic, which Math Academy says takes 30 to 45 minutes. Andy Matuschak logged his: 34 questions, 24 correct, about an hour.
The diagnostic places you on a knowledge graph. Each node is a topic and each arrow says one topic is a prerequisite of another. A course is about 300 topics and the whole curriculum is several thousand. The boundary between what you know and what you do not is your "knowledge frontier", and new lessons are drawn from it.
The bar is high and adults often land lower than they expected. Frank Hecker, who signed up because "my goal is to learn what an eigenvector is", called his first diagnostic "a brutal and dispiriting experience" and dropped down a course. When amun_dev wrote on Hacker News that the Linear Algebra diagnostic was overwhelming, Skycak replied in person and pointed him to the adult foundations courses.
Lessons, XP and spaced repetition
A topic is split into small steps. Each step opens with a worked example, followed by up to five practice problems, and two correct answers in a row move you on. Most answers are multiple choice. Your dashboard offers five tasks chosen by the algorithm and you pick among them.
Every task earns XP, and one XP is meant to equal about one minute of focused work. A timed quiz arrives roughly every 150 XP, with no looking back at the lessons. Anything you miss comes back as a review. A weekly league ranks you against other students by XP, and you can switch it off.
Reviews are scheduled by spaced repetition, with a twist that Skycak designed. Solving an advanced problem also exercises the simpler skills inside it, so the system gives those skills partial credit and skips reviews you no longer need. Hecker counted fewer than three reviews a day. Anki had overwhelmed him with them.
The courses for adults
The adult page describes Mathematical Foundations I, II and III as "especially crafted for adults who want to get up to speed or relearn math skills they have forgotten (from fractions through calculus)", with everything removed that is not a prerequisite for university math. The FAQ says a little less than an hour a day, five days a week, gets you through all three in a year. The adult page suggests a gentler 30 XP a day, four days a week.
Mathematics for Machine Learning follows. It covers linear algebra through eigenvectors and singular value decomposition, the multivariable chain rule ("essential for backpropagation", in the course description), and probability up to maximum likelihood. Skycak wrote in August 2024 that it had 242 topics and takes about 70 hours for someone who already knows single-variable calculus.
Reports from adult users
Jonathan Whitmore has a physics PhD and wanted his math back. His diagnostic put him 99% of the way through Foundations I and 76% through Foundations II. Two months and about 40 hours later he had finished II and started III. His verdict: "MathAcademy is extraordinary."
Hecker spent almost a month on Foundations II, starting at more than 90 XP a day, then did Foundations III between February 6 and mid-May 2025 at 50 or more. Of the 15 quizzes in Foundations II he had to retake five. He turned the leagues off after they started to stress him. In his update he called it "a wonderful service" and asked for one change: let the student choose which course comes next.
Kamil Staszewski wrote after 11,757 XP that he had missed two days since August 16, 2024. He fails most quizzes on the first try because of the clock, and he warns that the price is out of reach in many countries.
On Hacker News, cultofmetatron estimated that 8 to 9 hours a week takes you from Foundations I to the end of Mathematics for Machine Learning in a year and a half. modernerd spent about an hour a day, could not keep the pace, and blamed his own vague goal: "To stick with MathAcademy or any learning program it's better to have strong motivation in the form of a concrete goal you are working towards."
The recurring complaint is shallow understanding. Matuschak admires the scaffolding and the XP design, and then writes: "Theorems are often presented as things-to-be-learned, rather than things-to-be-understood." Oz Nova, who runs teachyourselfcs.com, wrote: "Explanations are minimal and motivating context virtually nonexistent." Michael Pershan, a math teacher, spent a month in Probability and Statistics, aced the quizzes, and reported: "I still don't know what a geometric distribution is."
Rendello finished Foundations I and II and described the result: "I feel like I learned 'Discrete Math', but in the sense that all the lessons were discrete and I couldn't draw connections between them." In the same comment he says he was "far better at math than I'd ever been".
Several users answer the complaint by pairing it with something else. rahimnathwani, with 3,722 XP of his own, suggests Math Academy for the exercises and 3Blue1Brown videos for the concepts.
Graded practice on a schedule works
Whitmore, Hecker and Rendello all report getting better at solving problems, and the critics mostly concede it. Pershan aced quizzes on material he says he did not understand.
This is the part Understand cannot do. It has no problem sets and schedules no reviews. If your goal is to compute a gradient by hand or factor a polynomial without thinking, you need hundreds of checked repetitions spread over months, and Math Academy is the best-documented way I know to get them. For that goal Math Academy is worth it at $49 a month. Whitmore, Hecker and Staszewski all say so, and I believe them.
A fixed path suits an exam
Math Academy is built around finishing. Lessons are passed, quizzes are timed, XP accumulates, and a completed course produces a transcript and a certificate. Its courses are accredited, and students taking one for credit get midterms and a final. The company started with schoolchildren who had a syllabus to cover, and for a student facing an exam I think this design is right.
The path is also not yours to change. Uri, at Atoms vs Bits, signed up for Linear Algebra because he loved it in college. After a few happy days his queue filled with trigonometry and calculus prerequisites as far ahead as he could see. He asked the team to let him study the parts of linear algebra that did not depend on them. They declined, and told him most of their customers are adults learning for fun or work. He quit: "instead of learning slightly sub-optimally, I'm now learning nothing at all."
Hecker wanted to know what an eigenvector is. By his own dates, his first months of daily work went to Foundations II and III.
Curiosity and its limits
An adult has no exam. What an adult usually has is a question: why backpropagation works, or what an eigenvector is doing in the paper open in the next tab. I think the question should set the order.
David Deutsch, in The Beginning of Infinity, follows Karl Popper in holding that knowledge grows by conjecture and criticism. In Popper's scheme a conjecture is always an attempt on a problem that somebody has. A curriculum supplies answers first and hopes the problems arrive later. Popper described the school he wanted to found in his autobiography, Unended Quest, as one "in which no unwanted answers to unasked questions would have to be listened to; in which one did not study for the sake of passing examinations."
Math Academy's reply is in its book, and it is a strong one. Skycak writes that math is a subject "where each advanced skill requires many simpler skills to be applied in complex ways", and that "maximal learning does not happen naturally as a result of maximizing other things like enjoyment, comfort, convenience, and ease of practice." He quotes his colleague Alex Smith on a mathematics PhD: "My biggest mistake when starting my doctoral research was taking a top-down approach." Someone who asks how backpropagation works without knowing what a derivative is will stall within minutes, and curiosity alone does not fix that.
I think both sides are right about different things. The prerequisite graph is real, and it tells you what a topic depends on. How much of it your particular question requires, and when you will care, the graph leaves open. Understanding why backpropagation works takes the chain rule and a picture of a function with many inputs. Deriving the gradients of a new architecture by hand takes the practice Math Academy sells. An adult who starts from the question finds out which of the two they are after, and arrives at the prerequisites with a reason to care. modernerd's lesson about concrete goals and Uri's exit both say that this reason is what keeps an adult going.
Where the question needs fluent calculation, curiosity loses and the schedule wins. Understanding that you could not use on a problem is also thinner than it feels, as I argued about audiobooks.
Understand by comparison
Understand is a conversation by voice, on any subject. You start it with a question, or with an article or PDF you saved. Ask about a dot product in the middle of a lesson on attention and the tutor answers there and carries on. Ask about mortgages and the call moves to a separate tutor for that subject.
It keeps a knowledge graph too, built the other way round. Math Academy's graph was drawn by experts before you arrived. Understand's is assembled from what you explain back in your own words, concept by concept, and you can read and correct it. Interest is recorded apart from understanding, so "skip the history" marks a topic as not for you without claiming you know it.
It has no problem sets, no spaced repetition and no curriculum. Its graph has no prerequisite arrows, so it cannot tell you what you are missing the way a diagnostic does. I described the mechanics and the missing evidence in my history of intelligent tutoring systems: nobody has run a trial of it.
Choosing between them
If you need the skill for a degree or a move into machine learning, subscribe to Math Academy, set 30 to 50 XP a day and expect a year or more.
If you have a question and no deadline, start with the question. Talk it through with a tutor, or read Nova's recommendation, a textbook that moves you.
Several of the people quoted here pair Math Academy with something that explains, and I would too. Rendello reached for ChatGPT to have a concept explained again. Do the daily XP, and when a lesson hands you the derivative of something you have never wanted to differentiate, take that to a conversation and ask what it is for.