I’m going to say what most EdTech founders are afraid to: your expensive AI grading software is probably useless. A complete waste of money. A solution in search of a problem that it solves badly.
I’ve seen it from every angle. As a founder who has built and sold two companies, one to Gusto and another to Gfycat. As an angel investor in over 200 startups, including some of the biggest names in AI like Anthropic, OpenAI, and Scale AI. And as someone who has spent countless hours mentoring founders in the EdTech space. They come to me with their shiny new AI grading tools, promising to save teachers time and provide "objective" feedback. And I have to be the one to tell them the truth.
It’s a lie. A convenient one, but a lie nonetheless.
I remember sitting in a demo a few months back. The founder, a bright-eyed and brilliant engineer, showed me how his platform could grade a hundred essays in under a minute. The dashboard was slick. The data visualizations were beautiful. But then I asked him to show me the feedback on a specific essay, one that was deliberately written to be creative and slightly unconventional. The AI had butchered it. It gave the essay a C-, complaining about "non-standard sentence structure" and "unusual topic progression." The "unusual topic progression" was a clever narrative device. The "non-standard sentence structure" was a stylistic choice that made the prose sing. The AI was a tin-eared robot, completely deaf to the nuance and artistry of the writing. It was penalizing originality.
That’s not an isolated incident. After testing dozens of these platforms, I’ve found that this is the norm. They are blunt instruments in a field that requires surgical precision. They are built on a fundamentally flawed premise: that education is a standardized, repeatable process that can be optimized for efficiency above all else. It’s not. Learning is messy, personal, and profoundly human.
The Core of the Problem: AI’s Blind Spots
The issue isn’t the technology itself. The large language models we have today are miracles of engineering. I should know; I’ve invested in the companies that build them. The problem is the application. It’s a classic case of a hammer seeing everything as a nail.
These grading tools are optimized for one thing: finding the “correct” answer. They work reasonably well for multiple-choice questions or simple math problems where there is a single, verifiable solution. But the moment you move into the realm of subjective, creative, or critical thinking, they fall apart. Writing, historical analysis, ethical debates, business case studies—these are the areas where real learning happens, and these are the areas where AI graders fail spectacularly.
Why? Because they lack true understanding. They are masters of pattern recognition, not comprehension. They can tell you if a sentence is grammatically correct, but they can’t tell you if it’s beautiful. They can check if you’ve mentioned certain keywords, but they can’t gauge the strength of your argument. They can spot plagiarism by comparing text to a massive database, but they can’t recognize the spark of a truly original idea.
I saw this firsthand with a startup I advised. They were building a tool to help grade code. On the surface, it seemed like a perfect application. Code is structured, logical. But the best programmers aren’t just logical; they’re creative. They find elegant, efficient, and sometimes downright weird solutions to problems. The AI grader, trained on a massive corpus of "standard" code, flagged these elegant solutions as "anomalies." It was teaching students to write boring, predictable code. It was training them to be average.
This creates a dangerous incentive structure in our schools. Students learn to write for the algorithm. They stuff their essays with keywords. They stick to simple, declarative sentences. They avoid risk. They optimize for the grade, not for learning. We’re training a generation of students to be robots, all in the name of a technology that was supposed to set us free.
A Better Way: My Framework for AI-Assisted Grading
So, should we just throw all the technology away? Go back to grading every single paper by hand, late into the night, with a red pen and a pot of coffee? Of course not. That’s a false dichotomy.
The goal isn’t to replace human teachers. It’s to augment them. To give them superpowers. The right question isn’t "How can AI grade for us?" but "How can AI help us become better graders?"
After years of thinking about this, and seeing what works and what doesn’t in my portfolio companies, I’ve developed a framework. It’s not as sexy as a fully automated dashboard, but it actually works. It leverages AI for what it’s good at, while keeping the human teacher firmly in control.
Step 1: AI for Triage, Not Final Judgment
Stop using AI to assign a final grade. It’s the single biggest mistake you can make. Instead, use it as an initial filter, a way to triage a large batch of assignments. When we’re looking at investment opportunities, we don’t have one person make a final decision from a list of 500 companies. We use processes and junior analysts to sort them into buckets: "Definitely No," "Maybe," and "High Priority."
Teachers can do the same with AI. You can use a simple model to scan assignments and flag them based on basic criteria:
- Completeness: Does the paper meet the word count? Are all sections of the prompt addressed?
- Clarity: Is the writing riddled with basic grammatical errors that make it hard to read?
- Similarity: Does the text show a high degree of similarity to other sources or other student submissions?
This first pass doesn’t assign a grade. It just organizes the work for the teacher. It tells you which papers need a quick look, and which ones require a deep, careful reading. It’s a massive time-saver, but it keeps the final judgment in the hands of the expert: the teacher.
Step 2: AI for Pattern Recognition Across the Class
An AI can read a hundred papers in a minute and tell you something a human might miss. It can tell you that 70% of the class misunderstood the concept of "irony," or that most students struggled to cite their sources correctly. This is incredibly valuable data.
Instead of just grading individual papers, the AI can provide a "State of the Class" report. It can highlight common misconceptions, frequent grammatical errors, or areas where the source material was confusing. The teacher can then use this information to adjust their next lesson. They can start the next class by saying, "I noticed many of you struggled with X. Let’s review that."
This shifts the focus from purely summative assessment (grading what’s done) to formative assessment (using data to improve future learning). It turns the AI from a judge into a diagnostic tool.
Step 3: AI as a Socratic Tutor, Not a Grader
This is perhaps the most powerful application. Instead of using AI to grade the final product, use it to help students improve their work before they submit it. Imagine an AI that doesn’t give you the answer, but asks you probing questions, just like a good tutor would.
- Student: "Is this a good thesis statement?"
- AI Tutor: "A strong thesis statement is usually debatable and specific. Does your statement present an argument that someone could reasonably disagree with? Could you make it more specific by mentioning the key evidence you plan to use?"
This is the kind of conversational, pedagogical AI that companies like Anthropic are pioneering. It’s an AI that acts as a coach, not a critic. It empowers students to think for themselves, to refine their arguments, and to take ownership of their work. It scales the best parts of one-on-one tutoring, making it available to every student.
Step 4: The Human-in-the-Loop is Non-Negotiable
Even with all these tools, the final call must rest with a human. The teacher’s intuition, their knowledge of the student, their ability to see the spark of a brilliant idea buried in a messy first draft—these are things that cannot be automated. The AI provides data points; the human provides judgment.
The grade is a conversation between the teacher and the student. The AI can inform that conversation, but it should never replace it. It can handle the grunt work, freeing up the teacher to focus on what they do best: teaching, mentoring, and inspiring.
Stop Chasing the Wrong Dream
I get it. The dream of a fully automated classroom is seductive. It promises a world of efficiency and scale. But it’s a hollow dream. It’s a vision of education that strips out the humanity, the creativity, and the critical thinking that we so desperately need.
Stop wasting your money on AI grading tools that don’t work. They are failing our students and burning out our teachers. They are a dead end.
The real opportunity is to build tools that empower teachers, not replace them. Tools that automate the drudgery, not the judgment. Tools that spark curiosity, not just check for correctness. That’s the future of EdTech. That’s where I’m putting my money. And that’s what will actually make a difference in the classroom.
Frequently Asked Questions
How can I apply this thinking to my own situation?
Start by identifying the core principle behind the opinion, not the specific example. Then ask yourself: does this principle apply to my context? If yes, test it in a small, low-risk way before going all in.
Do all experts agree with this view?
No, and that's fine. The best ideas in business are often contrarian. I share my perspective based on my experience and data, but I encourage you to seek out opposing viewpoints and form your own conclusions.
How has this view evolved over time?
My thinking on most topics has changed significantly over the years. Early in my career, I held many conventional views that experience proved wrong. I try to update my beliefs when the evidence changes.