Why is education so damn hard? Here’s an example.

Once, I was working on a national training program in partnership with a country in South America. We had about 10,000 students, all of whom were hard-working, diligent, smart, and focused on doing well at the program. A student came to my office hours and made me a proposal: “Jay, you’re a down-to-earth guy. Let’s take down these white people.” I didn’t know what to say. The next day, he was talking over the professor in class and posting across Slack channels: “Stop letting the Americans take over our country! Read about neocolonialism!” Most of all, it was the other students who wanted to take advantage of this unique opportunity who suffered. Eventually, we got him out of the program and learned that he was hired by the opposition government to undermine our program by creating chaos.

This kind of story will be familiar to most people who’ve worked with classrooms — at scale, education gets complex because outlier events create asymmetric downside risk. In the above story, it was an adult creating political trouble. But the same pattern appears when younger students have trouble at home, students have discipline issues, or adults run into issues with their financial situation. Almost every individual I’ve ever worked with is motivated to succeed,  but the number of potential outliers makes the risk of failure high and one scandal can create systemic risk.

This is why the natural state of educational programs at scale is a risk-averse bureaucracy. Arizona State University, often praised as the world’s most innovative university, isn’t great because it’s efficient — it’s great because it can deliver thousands of programs at once while maintaining talent density and avoiding systemic failure. Bureaucracy isn’t a bug — it’s a feature that protects downside risk by keeping humans in the loop for monitoring and redundancy.

The biggest mistake an EdTech company can make is thinking your students are the users. The second biggest mistake is thinking your teachers are the user. In truth, your key user is the bureaucracy. Clever was successful because they built a data layer that allowed the bureaucracy to optimize itself and plug in new technologies. LMSs were successful because they made content and course pathways visible to the bureaucracy.

AI infrastructure for bureaucracies is the only way to properly serve our learners

Learning outcomes: Measuring learning outcomes is a famously hard problem. Aristotle first called out that true learning can only be known through stable behavior, not test results. The best we can do to approximate a measure of learning outcomes is to be precise about the skills we want our learners to develop. AI learning infrastructure needs to absorb an institution’s context on their learner competencies and pathways to measure outcomes against their stated objectives.

ROI: The best educational institutions are mission-oriented, but institutions can only continue to serve their mission and drive impact with careful financial tracking. To invest in AI as a profit center, institutions need to compare human costs and financial costs to expected benefits beyond learning outcomes, such as program scalability. This can lead to more confident adoption and faster implementation at scale.

Safety: An app without institutional context will never know where to send safety flags, institutions’ cheating policies, or how to keep confidential data secure. This means that solving the AI safety and plagiarism problem requires a context engine that keeps AI-powered learning experiences within the bounds of an institution’s policies. This is how we can reach the holy grail of human-aligned AI that promotes leverage over cheating.

Jinso uses AI to make educational bureaucracies better and faster at serving their stakeholders

We’ve done this from day one. On the surface, Jinso is an AI tutoring platform that powers AI learning experiences using the Jinso Method. What makes us unique is our ability to turn bureaucracies into leverage, rather than drag. The humans in the loop — teachers and administrators — create data that the AI uses to keep improving our learning experience. This makes our friends in the bureaucracy indispensable partners and forces us to prioritize stability, functionality, trustworthiness, and local context over flashy demos.

• Instead of an app, Jinso is a modular platform
• Instead of supporting basic LTI, we support full LMS interoperability
• Instead of content, we provide AI-native learning experiences based on skills
• Instead of licenses, we use forward-deployed engineers to partner on implementation

Successfully implementing AI in education means learning from other companies that served bureaucracies. You don’t want to cut humans out of the loop — you want to understand why these humans are there in the first place. Consider some of the following examples.

AWS: Doesn’t prescribe how companies should deploy to the cloud. Instead, they provide a suite of tools designed for different members of the technical team to pick and choose according to their needs. Choose configurability over user-friendliness.

Palantir: Not just an enterprise AI product — a context engine that ingests institutional data from technical and human sources to plug into any application.

Salesforce: Understands that revenue is the tip of the spear for any company, and people across the bureaucracy want visibility for different reasons. Supports high configuration of visibility to different stakeholders’ needs and tastes.

Not everybody will care about this

If you’re a small private school, Jinso isn’t a fit. I recommend having your teacher create much more personalized experiences using ChatGPT. If you’re an individual who has the time and patience to build your own program, you also don’t need Jinso — I recommend building your own spaced repetition program with Anki in your desired field of study.

If, like me, you’re interested in how large programs can maintain quality and safety at scale, Jinso is built for you. We’ve worked with learners in dozens of countries, from young students to working adults. We don’t promise flashy demos or safety theater — what we promise is sustained, safe, and effective execution that cares about how you operate and helps you maintain what makes you great.

The post Education needs infrastructure, not apps first appeared on Jay Lee-Gopalan.