What this covers
Patrick talks with Joe Liemandt about Alpha School and about the mature-software business he ran before it. The education argument starts from a diagnosis: time-based schooling moves students forward with gaps in their foundations, like Swiss cheese, and the gaps collapse at algebra. The fix is hole-filling along a Prerequisite Hierarchy, so a sixth grader missing fourth-grade material gets fourth-grade lessons. Liemandt leans on Benjamin Bloom's tutoring research, spaced practice over cramming, and content tuned so students succeed 80 to 85 percent of the time, which keeps them in the Zone of proximal development.
The middle of the conversation is the most contested. Liemandt says the median student gains one point on the 300-point MAP test across four years of high school, and that the one-teacher classroom is the problem, not teachers. He says EdTech works mainly for the 5 to 10 percent of students who are self-motivated. His motivation tools include "Alpha bucks" tied to an investment account with an early-withdrawal penalty, and using money to break fixed beliefs about ability. He also describes parents rejecting the "2x learning" pitch until it became two hours and time back. The talk touches parental fear of struggle and teaching goal-setting method from second grade.
The last stretch returns to software. Liemandt describes the Melting Ice Cube approach to legacy products, with customer success as the retention engine, and predicts pure transactional SaaS will lag AI-native companies over two decades. He argues that education needs business models because nonprofit funding limits scale, and that a founder must push each non-negotiable to its extreme, since an 85 percent version of the system fails.
Liemandt argues that AI tutoring built on mastery-based learning science and a closed measurement loop can compress K-12 academics into about two hours a day, and that the binding constraint is student motivation (giving kids their time back), not the software itself.
- Mastery-based, prerequisite-aware AI lessons let students learn roughly 2x to 10x faster than time-based classroom instruction, with measurable gains on standardized tests like MAP.
- EdTech alone has failed because it is necessary but not sufficient; motivation is about 90% of the solution, and giving kids their time back is the strongest motivator.
- Scaling depends on a closed feedback loop from student behavior to content generation to test outcomes, plus entrepreneurs building schools and businesses on a shared platform.
Motivation, not content, is the binding constraint, supplied by giving time back, games, and incentives.
- The most powerful motivator for students is giving them their time back, meaning finishing academics in two hours so they can spend four hours on engaging activities, and other incentives like stickers or money only work once this is in place.
“by far the number one motivator for kids... is give kids their time back.”
- EdTech succeeds mainly for roughly 5 to 10 percent of students, who are self-motivated, high-conscientiousness, often high-IQ learners who love academics for their own sake, so the remaining 90 percent of the market requires solving motivation.
“EdTech has 5 or 10% of kids who do well with it.”
- Educational technology has failed to raise test scores because it addresses only the content and difficulty-matching part of learning, while motivation and engagement constitute about 90 percent of the solution and EdTech about 10 percent.
“Motivation is 90% of the solution. The EdTech is 10%.”
- Kids want to learn as quickly as possible when they are given back their time, so learning speed is driven by the incentive structure rather than innate desire.
“Kids want to learn as quickly as they can, assuming you give them their time back.”
- Schools should be designed so kids love school enough to prefer it over vacation, and this commitment, held as a non-negotiable standard, is the foundation of Alpha's design.
“kids must love school. That's just a crazy thought that most people do not believe you.”
- Children are engaged by challenges that feel hard and risky but are supported by a caring adult, and parents' fear of struggle undermines the very workshops that make kids love school, so high standards are the key to happiness.
“The kids want to go do hard things. They don't want to struggle and fail, but they want to struggle and fail on their road to success.”
- Educators say a student must first be motivated and then placed in lessons of the right difficulty, so EdTech handles only about 10 percent of the solution, and Liemandt argues that Alpha is built around solving motivation, which is the other 90 percent.
“Motivation is 90% of the solution. The EdTech is 10%.”
- A game resembling Candy Crush can teach math fact fluency in a way that kids voluntarily play, and this fluency will move math capability more than most interventions, showing that engagement is the scalable route.
“think of a game Candy Crush, but it's teaching math fluency”
- Kids will voluntarily practice academics when the reward is in-game advantage such as extra health, as one child said she would do ten hours of academic training for a game benefit, so video game incentives can be redirected to learning.
“This kid literally just was, I would do 10 hours of academics.”
- Students complete 25-minute subject sessions called Pomodoros and earn closed rings only by reaching mastery at about 80 percent accuracy or higher, so guessing does not count, and unlocking the afternoon depends on closing these rings.
“So you have to have 80% accuracy. You did it to mastery.”
- Showing students a waste meter that quantifies the percent of their time spent on low-value activity creates a self-directed desire to become faster learners, reducing the time needed to complete the two-hour session.
“So we literally are saying, you're wasting 50% of your time. And the kid's thinking, whoa, coach me up.”
- Parents would do almost anything to have their child experience a breakthrough in which they realize they are limitless, and a modest incentive like 1000 dollars can trigger this shift in self-belief, which matters more than the money itself.
“I now realize I can do anything. There is no academic height.”
- A video-game school that offered two-hour learning followed by afternoon gaming was hard to sell to parents, but the same model framed as after-school tutoring with tournament rewards was attractive.
“But there's parents are saying, 'I'm not sending my kid to video game school.' That's a hard push.”
- Students earn Alpha bucks for academics that can fund passion projects or a real investment account with a 50 percent early-withdrawal penalty, because money that is earned rather than given teaches saving and avoids the loss aversion problem.
“the key if you're going to teach a kid to invest is make sure it's real money in theirs and it's something they earned, not given.”
- There are about 10 million American mothers who would pay 100 dollars per month for an academic video game that drives their kid to top 10 percent academically, so educational games can be highly monetizable.
“There's 10 million moms in America who would pay 100 bucks a month for their kid to love academics and drive it up.”
AI-delivered instruction frees human guides and teachers to become motivators and mentors building life skills.
- The best adults in a school are those who combine high support with high standards, and the best measure of their success is whether students see them as a teacher who transformed their life, so guides should be hired on motivation rather than academic instruction skill.
“You are looking for individuals who can hold high standards, but also high support.”
- Teachers and adults are valuable and should be retained in a new role focused on motivation and mentoring, while the problem to solve is the format where a teacher must deliver academic instruction to a whole class.
“Teachers are great. Teacher in front of the classroom is the problem. The model is the issue.”
- Highly successful adults tend to over-invest in ambitious teenagers, so students pursuing projects can recruit adult mentors who help them far beyond what a school can staff.
“super successful adults over invest in ambitious teens”
- Alpha pays guides a minimum of 100,000 dollars in the Austin market, roughly double typical teacher pay, because the economics of schooling are driven by teacher pay and class size, and the lower grades can run at a five-to-one student-to-guide ratio.
“Our minimum pay for a teacher is $100,000 in the Austin market. That's basically double the pay.”
- Bloom's Two Sigma paper showed that students with individualized tutors perform far better than classroom peers, but the finding was dismissed as infeasible at scale, which Liemandt sees as the central constraint that AI now removes.
“One of the first papers, there's Bloom's Two Sigma, a seminal paper.”
- Adults in Alpha buildings do not deliver academic instruction; AI does, so adults are guides and coaches responsible for motivation, belief in students, and life skills, and Liemandt argues that the robot-tutor fear is a misconception.
“if you are an adult in our buildings, you do not do academic teaching.”
- Because guides do not teach academics, Alpha hires former NBA coaches, NFL players, and MLB players, since they excel at holding high standards and motivating students.
“we actually have ex-NBA coaches, ex-NFL players, ex-MLB players.”
- Students retain about one to five facts or concepts per hour of a Stanford lecture, while Alpha's learning engine delivers about 20 to 40 per hour, and a planned TikTok-style interface aims for 100.
“if you sit in a Stanford lecture, you retain about one to five facts or ideas or concepts per hour.”
- The remaining four hours are project-based, team-based workshops teaching leadership, teamwork, entrepreneurship, financial literacy, public speaking, and grit, which Liemandt argues are the skills parents want beyond academics and which were historically taught mainly through sports.
“And those are big project-based workshops that are super fun for the kids and team-based, not screen time.”
- Timeback is focused on K-12, so adult self-directed learning on the same engine would need a separate product, though he expects the approach to become standard for everyone within five to ten years.
“We're focused on K through 12, and so there needs to be somebody else who does that for the adults.”
Students should advance by mastered content and bottom-up prerequisite repair, not by age-based grade level.
- Students who spread vocabulary practice over 10 minutes a day retain material a week later, while students who cram the night before forget most of it, so cramming gives a short-term score but no retention.
“the cram kids don't remember anything. And the people who've done it 10 minutes a day all remember it”
- A student's age-appropriate grade and the content level they have mastered are different things, so a sixth grader missing fourth grade foundations should be given fourth grade lessons rather than sixth grade material.
“age grade and knowledge grade are very different things”
- When students struggle with advanced material such as algebra or fractions, the real cause is usually a gap in prerequisite knowledge like division or multiplication, so remediating earlier material fixes the later difficulty.
“The problem's not the algebra, it's the prior knowledge.”
- Students should be held to a mastery standard above 90 percent, since the 70 percent passing bar typical in standard schools leaves 30 percent gaps that compound, and mastery standards embedded in existing apps are too low to produce Alpha-level results.
“So our standard is over 90.”
- Time-based schooling, where students advance without full mastery, creates holes in foundations like Swiss cheese, which eventually collapse at algebra, so the fix is to backfill missing knowledge rather than push students forward.
“It's like you're building a foundation with all these holes.”
- Most knowledge is hierarchical, with algebra built on fraction manipulation and fractions built on multiplication and division, so students must master foundations bottom-up rather than moving forward on partial understanding.
“Most of knowledge is hierarchical where it's based on foundation.”
- Alpha's K through 6 building in Austin earns more perfect 100 scores on the Texas STAAR than an entire school district of 100,000 students, which Liemandt attributes to a mastery-based rather than time-based model.
“gets more hundreds on the Texas STAAR than a school district of 100,000 kids.”
The industrial one-teacher classroom model, not teachers themselves, blocks individualized learning.
- The 200-year-old industrialized model of one teacher in front of a classroom of 20 to 30 students was adopted to scale education globally, and it is structurally inferior to individualized instruction, so the model rather than teachers is the problem.
“fundamentally, 200 years ago, when we decided everybody needed education globally, we created the industrialized model”
- On the 300-point MAP scale, the median American student increases by only one point over four years of high school, indicating essentially no learning, and the flattening starts in middle school.
“the median American goes up one point in four years, one point on a 300-point scale. Not learning anything.”
- Learning science has documented for about 40 years that methods such as individualized tutoring, mastery learning, spaced repetition and cognitive load management can make kids learn 2 to 10 times faster, but these findings were never implemented because individualized instruction was not scalable.
“they started writing papers 40 years ago that were basically, the teacher in front of the classroom's the worst way to teach kids”
- In a traditional classroom, teacher quality, student motivation, prior knowledge and classroom environment dominate outcomes, so it is impossible to isolate the effect of instruction, whereas AI tutoring enables precise instruction and measurement.
“if you take the AI out, your effect is dominated by the classroom and you don't know”
Business models and AI-native software, not donations, are how education scales beyond physical campuses.
- Pure transactional SaaS will produce dramatically worse returns than AI-native software companies over the next two decades, because AI lets software change how a business works and cuts the cost of replacing legacy systems.
“Pure transactional SaaS I believe is just going to be dramatically worse than the big winners, the AI-native ones.”
- Education is a trillion-dollar market where a great product can attract paying customers, but education is structurally constrained by nonprofit, donation-funded models that prevent sustainable scaling, so capitalism and business models are needed to fund expansion.
“Great products, parents will absolutely pay for great products. Societies pay for great education.”
- Gifted and talented students who love academics for their own sake may learn 3x to 5x faster than the typical 2x, so Liemandt envisions 500 GT schools built on the platform to unleash that population.
“Those kids aren't learning 2x in 2 hours. They're learning 3x, 4x, 5x.”
- The Timeback engine is not yet ready to scale because AI still produces hallucinations, and student screenshots of bizarre outputs illustrate the reliability gap that must be closed before broad rollout.
“And the engine, the Timeback engine, we still have hallucinations, right?”
- Liemandt says more than $100 million has been invested in video game designers building an AAA-quality game on the learning engine, betting that parents will pay for educational games that children want to play, and that video games can deliver the motivation layer at scale.
“We've invested over $100 million in getting some of the best video game designers in the world.”
- All existing educational content, including textbooks, lesson plans, and tests, is obsolete because generative AI can produce a personalized lesson matched to a student's knowledge graph and interest graph.
“all educational content is obsolete. Every textbook, every lesson plan, every test, all of it is obsolete.”
- A proposed Texas network of micro-schools, funded by new school choice dollars, would use the Timeback platform for two hours of academics and then have local coaches run athletics in the afternoon, creating a model for D1 athletes that could enroll tens of thousands of kids.
“Texas just passed school choice billion dollars.”
- Alpha's results come from a limited, selected, high-resource group, so the central skeptical objection is that the model has not been proven at scale, and claims about billions of students in ten years cannot yet be supported by current data.
“There's nothing in Alpha today that would say—there's a billion kids 10 years from now—that's going to work.”
- Game developers have the expertise to engage kids, so an educational video game built on the Timeback learning engine could teach a grade level of math with the same engagement that commercial games achieve, and parents might pay for it, enabling scale beyond physical campuses.
“the people who know how to engage kids more than anything is video game developers.”
- Timeback, which teaches kids ten times faster, is the single best product Liemandt has built, because no other software has ever enabled kids to learn ten times faster and its economic value could be very large.
“There's zero doubt in my mind that Timeback as a product is the single best product I've ever built.”
Third-party testing, visible results, and trusted guides must overcome skeptical parents' outdated assumptions.
- Parents initially doubt the two-hour model, but when January MAP results show twice the expected learning, skeptical parents are convinced, so third-party standardized testing is used to validate outcomes that parents would not otherwise believe.
“the January MAP test come back and it shows they're learning twice as much in two hours a day.”
- The Alpha team takes ideas from learning science texts, such as a chapter by Dan Schwartz, generates lessons, measures engagement and test outcomes, and loops the results back to refine the approach.
“let's take Dan Schwartz, chapter 10 and put it in.”
- A student scoring 740 on the math SAT was sent back to third-grade math to build multiplication and division fluency, and her score rose to 790 after she stopped making careless errors caused by overloaded working memory.
“you can solve harder math problems if you know your multiplication tables and division tables.”
- A second-grade guide taught students Atomic Habits concepts and used incremental walk-run progression so that by December every second grader could run a 5K in under 35 minutes, showing that difficult goals become reachable through systematic incremental practice.
“all the second graders can run a 5K in less than 35 minutes.”
- Parents who see video of kids climbing a 40-foot rock wall or running a 5K realize that challenges look dangerous from outside but are enjoyable inside a safe, supported environment, which makes the approach believable.
“when you actually see the kids go through it, they love it, but it's hard.”
- The Timeback platform will launch a homeschool version next year that can deliver 2x learning in two hours, though current homeschool testing has not yet matched Alpha results without added motivational models.
“So I believe next year we will have a homeschool version that somebody will be able to buy.”
- Parents naturally want children educated the way they were, which is why even a founder's own family resisted his wife's new school, and this familiarity bias is the normal parental reaction to educational innovation.
“every parent wants to educate the kid the way they were.”
- Schooling is a complex bundle of experiences and peers beyond academics, so even a student who finds lectures unhelpful can benefit from Stanford's environment, and Liemandt says his daughter now uses a math app rather than a lecture for learning.
“academics is one piece of a very complex bundle.”
- Parents believe a child must stay at grade level no matter what, which is the opposite of what learning science and tutors do, so Liemandt says persuading parents to accept backfilling is half of his job as principal.
“They are parents have been taught, my kid needs to be at grade level no matter what.”
- Alpha uses guides and coaches instead of teachers because adults in the building no longer do academic teaching, and the benchmark is whether the guide is one of the one or two adults who transformed the student's life, as measured by an eight-week survey.
“Is your guide that for you?”
- Three beliefs would change education if every parent held them: kids must love school, kids can crush academics in two hours a day, and high standards are the key to a child's happiness.
“your kid must love school... your kid can crush their academics in two hours a day... high standards is the key to your child's happiness”
- The name Incept reflects two meanings: putting thoughts into a child's mind, as in the film, and the historical sense of being qualified to teach, which implies that educators must be trusted generators of what goes into students' minds.
“Incept sort of has two definitions.”
- Parents' views of education are shaped by two centuries of old assumptions that are no longer true, so the next phase of Liemandt's work must educate parents about what is now possible.
“Educating parents about what is possible. My new role is I have to go out and do that.”
- The kindness that sustains Alpha is students' willingness to buy into a startup school and trust the adults, which Liemandt sees as the core that makes the enterprise work.
“their willingness to be part of it is the thing when I think of gratitude and kindness, that's the core that makes this all happen.”
Generative AI personalized mastery lessons engage students, improve retention, and fit each student's capacity.
- Generative AI can create personalized lessons that combine a student's knowledge graph, interest graph and curriculum, such as teaching fractions through fashion or World War I through Taylor Swift, which makes content engaging and improves retention.
“this is the generative part of AI that really matters is all content, all educational content is obsolete.”
- Around 90 percent of general ChatGPT use by students is for cheating, so deploying a general chatbot to schools is harmful, and Alpha's AI deliberately omits chat functionality in favor of lesson generation and coaching.
“90% of the use of ChatGPT is to cheat.”
- Working memory holds a limited number of chunks, and the number of practice repetitions needed to move material into long-term memory varies by student, with gifted students needing fewer reps, so lessons should be sized to each student's capacity.
“Every person has a certain number of chunks that you can work on or registers on a CPU”
- Students who lack fluency in multiplication and division tables waste working memory on basic calculation, which causes careless errors in advanced math, so memorizing basic facts can raise higher-level math scores.
“you can solve harder math problems if you know your multiplication tables and division tables.”
- Just as the microscope, telescope and analytical balance enabled biology, physics and chemistry to become rigorous sciences, generative AI is an instrument that allows precise measurement and instruction, letting learning science finally move out of its wilderness period.
“there was an invention of an instrument, the microscope for biology and the telescope for physics and the analytical balance for chemistry”
- Learning content should be set so students succeed about 80 to 85 percent of the time, which keeps them in the zone of proximal development and engaged, as game developers do.
“knowledge at 80 to 85% accuracy, which is sort of the zone of proximal development”
- The Alpha AI tutoring system currently costs about 10,000 dollars per student, which Liemandt expects to fall below 1,000 dollars on a tablet within five years.
“there's going to be a tablet that's less than $1,000 that is going to teach every kid on this planet everything they need to know in two hours a day.”
AI tutoring in two hours outpaces classroom learning, freeing the rest of the day for development.
- On third-party MAP tests administered in August and January, Alpha students show about twice the expected learning growth in two hours a day compared with students in traditional six-hour classroom days, and growth is off the charts relative to other US schools.
“the January MAP test come back and it shows they're learning twice as much in two hours a day.”
- Students at Alpha spend about two hours a day with an AI tutor and learn academics at roughly 2x to 10x the rate of classroom instruction, so academics can be finished in two hours and the rest of the day freed for other development.
“kids go in, spend two hours a day with the AI tutor with screen time and all the things around that, we obviously consider it very good screen time, and they learn 10 times faster.”
- Alpha students earned all fives on AP World History after a maximum of 75 hours of AP study, challenging the common assumption that AP courses require hundreds of hours.
“They all get fives and they spend 75 hours max to do AP.”
- Teaching an entire grade level of material in one subject takes about 20 to 30 hours with the Alpha system, compared with the 180 school days of hour-long classes plus homework in standard schooling, so a kid a year behind is only about 20 to 30 hours behind.
“You can teach an entire grade level of material for one subject in about 20 to 30 hours.”
Schools should teach goal-achieving methods and broader development beyond academic test scores.
- Telling children to believe they can do anything is insufficient; schools should teach them the concrete method of achieving goals, such as incremental practice in the style of Atomic Habits, starting in second grade.
“We teach kindergarten, first grade. Teach a lot of growth mindset. You should think you can do anything. And she's like, we got to start teaching them in second grade how to do anything.”
- In high school, students spend their afternoons for four years on a self-directed passion project, with the school providing outside adult support and market feedback rather than in-house coaches, since no staff member can cover such varied interests.
“we have what we call Alpha X project, which is your four year super passion project”
- Early Alpha pitched academic gains of 2x learning to parents and was rejected, and only after reframing to two-hour learning and time back did parents engage, showing that most parents value other aspects of school more than academic achievement.
“one of the things that I did not realize going in is how few parents actually care about academics.”
- Schooling is a complex bundle of social environment, peers, networks and experiences, so even a top academic institution like Stanford delivers value beyond lectures, which is why academic effectiveness alone should not be the sole measure of education.
“academics is one piece of a very complex bundle.”
Mature software businesses should generate cash by retaining customers, leaner operations, and cutting growth costs.
- AI makes it much easier to replace long-tail legacy systems that were previously impossible to swap out, so mature SaaS businesses melt faster, though the disagreement is about the rate of melt rather than whether AI accelerates it.
“AI allows you to replace legacy systems much easier than you could pre-AI. No one disagrees with that. It's just your rate of melt.”
- For mature software businesses, cash flow comes from retaining the installed base by calling customers, scheduling quarterly success reviews, and fixing their problems, rather than chasing new prospects as growth-stage companies do.
“The number one is growth versus install base. In a growth company, you care about the new customers”
- Enterprise software must show customers a return on their investment, not only offer a good product, because a product that does not deliver return is a failing product no matter its awards.
“If General Electric doesn't get a return on its investment, your product sucks.”
- Hypergrowth cost structures and culture are very expensive, so when a company is no longer a growth play, cutting the growth-oriented cost base and culture generates cash flow, even at the cost of some employees who resist the change.
“Growth is expensive. A way to do it is, if you're trying to be a hypergrowth company, your cost structure and your culture, it's just very expensive.”
- Many SaaS unicorns loaded with debt during growth plays now have near-zero equity and are handed to private credit lenders, and an operator can acquire them for about a dollar and split cash flows, representing a large opportunity worth about 20 billion dollars in such companies.
“there's $20 billion of those companies out there on the books.”
- Legacy software with declining terminal value should be managed as a melting ice cube, extracting cash flows and investing only where it extends the useful life, because a discounted cash flow valuation can still be positive even if terminal value falls to zero.
“you're looking at your DCF, you're pulling the cash out and if it goes to zero”
- Customer success that includes quarterly calls and shared metrics restores retention among legacy enterprise customers, which Liemandt says is the core of the mature-software playbook, with upsell following from solving customers' problems.
“And when you do that, all of a sudden you get retention, retention of the installed base.”
- ESW's acquisition model is to take mature software companies that have achieved product-market fit, run them with lean operating discipline, and generate more cash flow, which was the strategy used long before software holding companies became popular.
“Growth is expensive.”
- Jack Welch told Liemandt that if General Electric did not get a return on its investment, the product was bad, which pushed Trilogy to build customer success and value demonstration rather than just selling a product of the year.
“If General Electric doesn't get a return on its investment, your product sucks.”
Mastery, effort, motivation, and challenge can change students' self-belief and their approach to feedback.
- Offering students 100 dollars for a 100 percent score on any grade's Texas STAAR, with prior-grade tests taken first and mastery-based AI lessons used to prepare, cost roughly 400 dollars per student and both fills learning holes and changes students' self-belief.
“And then you do sixth grade. And for literally $400, we can do two really important things.”
- In a mastery-based system nearly everyone can score 100 percent on a state test, because success depends on effort rather than innate intelligence, and breaking the belief that ability is fixed changes how students see themselves.
“In a mastery-based model, everybody can. It's just a matter of effort. It's not about IQ, it's just about effort.”
- Students should review their own test errors like athletes watch game film, sorting each miss into a knowledge gap or a careless mistake, which turns academic feedback into a self-improvement habit.
“We call it watching game film. Great athletes are like, yeah, I'm gonna go watch the game.”
- Learning to seek and receive critical feedback without defensiveness is one of the most important life skills to teach, and Alpha tests it from kindergarten (can you receive feedback without crying) through eighth grade (do you actively seek it).
“receiving feedback is one of our most important life skills to teach”
- Liemandt offered his daughter $1,000 to reach the top 1 percent after she settled for the top 10 percent, and her shift from believing she was not as capable as her sister to believing she could learn anything shows how a small reward can change a child's self-belief.
“I now realize I can do anything. There is no academic height.”
- Kids want to do hard things and grow from struggle when supported by a caring adult, and kindergarteners climbing a 40-foot rock wall or second graders training for a 5K show that growth mindset becomes real only after the child attempts something difficult.
“The kids want to go do hard things. They don't want to struggle and fail, but they want to struggle and fail on their road to success.”
- Financial rewards for academic achievement can be used deliberately to break fixed self-beliefs, such as a student concluding she is not as capable as her sister, and the belief change persists after the money is spent.
“I now realize I can do anything.”
- When Alpha's fifth graders were asked what would make them love school, they proposed less school, and a negotiated deal of two rigorous hours of apps in exchange for four hours of enjoyable workshops was accepted.
“If you really engaged with the apps for two hours and then you had four hours of the most awesome stuff”
School models should be differentiated and built to uncompromising standards, since partial implementation fails.
- Once the Timeback academic engine is proven, entrepreneurs should build differentiated schools on top of it, such as micro-schools for athletes, GT schools, or finance-focused schools, because parents want different things and one size will not fit all.
“entrepreneurs need to build new schools. And this two hour platform will cover the two hours for you.”
- Public school leaders face a far harder problem than a private operator, because they cannot control the product, must serve parents with diametrically opposed views about the purpose of education, and must rebuild the whole school to adopt the platform.
“Public schools, wow. They have a 10 times harder job than me”
- Trilogy recruited top Ivy League and elite university graduates to Austin by promising a brutal 100-day program of 100-hour weeks, and this high standard won 70 percent of the candidates compared with 30 percent for Microsoft.
“Against Trilogy it was Microsoft only got 30% and we won 70%, and the reason was we made it harder.”
- Design decisions in a system interact so that partial implementation of a standard, such as kids loving school, fails, so the founder must push each non-negotiable to its extreme, because an 85 percent version of the system does not work.
“If 85% work, I do that. It doesn't. I'm pushing it to where it needs to be for it to work.”
An AI vision model monitors student screens to coach them toward more efficient learning.
- Alpha's advantage is a closed loop in which its team generates a lesson, students use it, the system measures time and engagement, and standardized test results confirm learning, a loop no other school or edtech company has, which explains the outcome gap.
“We're basically the only place that has that closed loop, which is why we're able to get these crazy awesome outcomes and other people aren't.”
- Alpha streams student screen video to an AI vision model that detects anti-patterns such as skipping explainers, scrolling articles, or searching for answers, updates a waste meter, and coaches the student toward more efficient learning.
“the video stream itself is actually being fed to an AI model sitting there looking for anti-patterns.”
- The vision model was initially trained by human annotators who watched student screen videos overnight and labeled learning anti-patterns, which became a reinforcement learning with human feedback loop that has since been mostly automated.
“literally just humans, when you talk about human annotation, just watching videos.”
Liemandt's software-business background and personal drive to push limits shape the venture.
- Liemandt as a teenager aimed for the minimum academic score that still met goals, hitting 89.5 to round to an A, and this pattern of minimal-but-sufficient effort later became his drive to push things to the limit.
“I was always about what's the minimum academic score I can get but still hit goals.”
- Software businesses are fundamentally HR businesses because people are their largest cost and biggest source of return, so talent and management quality determine outcomes more than any other lever.
“software businesses are HR. They are HR businesses. It's your biggest cost structure.”
- Trilogy's first product was an expert-system configurator called Sales Builder that encoded engineering and manufacturing rules into laptops for salespeople, preventing invalid configurations that cost companies millions to fix on the factory floor.
“what our system would do is called Sales Builder.”
- After organic product builds in the 1990s, Trilogy and ESW shifted after 2000 to acquiring software companies, generating enough cash flow to fund later ventures like Alpha School.
“after 2000, we started buying companies, not just builds.”
Simple, memorable strategy and selective values expose whether teams agree and attract the right people.
- Strategy should be distilled to three lines of three words each, since long strategy documents let people claim alignment with any one line, and a simple clear message exposes whether teams actually agree.
“You're allowed three lines, three words each.”
- Highly educated employees often suffer from smart person disease, which means they overcomplicate strategy and are overly impressed by elaborate documents, so leaders must communicate in simple, memorable form.
“Joe, smart person disease. You have smart person disease.”
- Effective values and marketing must act as a magnet that attracts the right people and repels others, so a core value that no one would oppose, such as integrity, is wasted and a list item should be included only if someone would take the opposite position.
“If someone won't say the opposite, it shouldn't be on the list because then it's a platitude.”