'''
The Emperor Has No Clothes: Why Your AI Grading Tool is a Million-Dollar Mistake
I’m going to say what most EdTech founders and VCs are terrified to admit: your AI grading software is probably a complete waste of money. There, I said it. A multi-million dollar industry built on a foundation of sand.
I’ve seen it from every angle. As a founder who 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 personally tested dozens of these so-called "revolutionary" AI grading platforms. They’re mostly garbage.
It started a few years ago. I was mentoring a startup in the EdTech space. They were building an AI-powered tool to automate essay grading for high school English classes. The pitch was seductive. Save teachers countless hours, provide instant feedback to students, standardize grading across the board. It sounded like a dream. They had a slick demo, a team of PhDs, and a waitlist of schools ready to throw money at them. I almost invested.
But then I asked to see the data. Not the curated, cherry-picked examples from their pitch deck. I wanted the raw, messy, real-world stuff. I gave them a set of essays from a 10th-grade literature class. The topic was the use of symbolism in The Great Gatsby. The essays were…well, they were written by 10th graders. Some were brilliant, some were a mess, and most were somewhere in between.
The AI’s grades were all over the place. It gave an A+ to a beautifully written but completely surface-level essay that just re-summarized the plot. It slapped a C- on a messy, grammatically flawed essay that had a truly original, mind-blowing insight about the green light. The AI completely missed the spark of genius because it was buried in imperfect prose. The tool was rewarding conformity and punishing creativity. It was a disaster.
The Nuance Blindspot
Here’s the fundamental problem: current AI is terrible at understanding nuance. It’s a pattern-matching machine. It can check for grammar, sentence structure, and keyword usage. It can even be trained to recognize certain rhetorical devices. But it cannot, in any meaningful way, appreciate subtlety, irony, or a truly original thought that breaks from the expected pattern. It can’t tell the difference between a student who is genuinely struggling and one who is brilliantly experimenting with form.
I saw this again and again. Tools that would penalize a student for using a sophisticated, but less common, synonym. Platforms that would get confused by complex sentence structures and mark them as run-ons. And the worst offenders, the ones that would literally just count the number of times a student mentioned the "key themes" from the prompt, rewarding keyword stuffing over genuine understanding.
We’re training a generation of students to write for the algorithm. To churn out safe, predictable, five-paragraph essays that check all the boxes but have zero soul. We are optimizing for mediocrity.
I get the appeal. I really do. Teachers are overworked and underpaid. The promise of offloading the soul-crushing burden of grading hundreds of essays is incredibly tempting. But the solution isn’t to abdicate our responsibility to a faulty algorithm. The solution is to use AI as a tool, not a replacement.
A Better Way: The Co-Pilot Framework
After that eye-opening experience, I started developing my own framework for using AI in education. I call it the "Co-Pilot Framework," and it’s built on a simple principle: AI should be the assistant, not the master. It’s about augmenting human intelligence, not replacing it.
Here’s how it works:
First Pass for the Basics (The AI’s Job): Use a simple AI tool (honestly, even the grammar check in Google Docs is a good start) to do a first pass on the mechanics. Spelling, grammar, punctuation, basic sentence structure. This is what AI is good at. It cleans up the low-level stuff, so the human grader can focus on what matters.
Identifying Potential (The AI’s Other Job): This is where it gets interesting. Instead of asking the AI to assign a grade, I ask it to flag things for me. I use custom prompts to ask it to find the "most original sentence" or the "most surprising connection." I ask it to identify passages that deviate from the standard interpretation. I’m not asking it for a judgment, I’m asking it for a signal. I’m using it as a discovery engine to point me to the interesting parts of the essay I might have missed.
Deep Reading and Feedback (The Human’s Job): This is the part that can never be automated. This is where the teacher, the expert, the human, actually reads the essay for its ideas. They engage with the student’s argument, appreciate their unique voice, and provide the kind of nuanced, insightful feedback that actually helps a student grow. They can now do this more effectively because they’re not bogged down in correcting comma splices.
The Final Grade (The Human’s Final Say): The final grade is, and must always be, assigned by the human. It’s a holistic judgment based on the quality of the ideas, the strength of the argument, the originality of the voice, and the effort of the student. The AI’s input is just one data point among many.
I’ve used this framework with schools I advise, and the results have been incredible. Teachers are happier because they’re spending their time on meaningful interaction with students, not on rote correction. And students are writing better, more creative, more interesting essays because they know a human is on the other side, ready to engage with their ideas. For more on how I think about building products that work, you can check out my post on the art of the pivot.
Stop Chasing the Shiny Object
Look, I’m not an AI pessimist. I’ve invested in some of the most important AI companies on the planet. I believe this technology has the power to transform our world for the better. But I’m also a realist. I’ve seen too many founders and investors get blinded by the hype, chasing the dream of a fully automated solution without stopping to ask if that’s what we really want or need.
In education, we don’t need more automation. We need more connection. We need more mentorship. We need more of the messy, inefficient, beautiful human stuff that actually inspires a kid to fall in love with learning. If you ’re a school administrator or a teacher, I’m begging you, don’t get seduced by the slick sales pitches. Run your own tests. Give the AI your most creative and your most challenging students’ work and see what it does. The answer will likely scare you.
And if you’re a founder in this space? Stop trying to build a magic black box to replace teachers. It’s a fool’s errand and, frankly, an insult to the profession. Instead, build tools that empower them. Build co-pilots. That’s a mission worth pursuing. It’s a company worth building. And it’s the kind of company I would actually invest in. You can read more about my philosophy on this in my post about my investment thesis.
The Real Takeaway
This isn’t just about grading software. It’s about our entire approach to technology. We have to move past the naive techno-optimism that believes every problem can be solved with a clever enough algorithm.
Some things are meant to be inefficient. Mentorship is inefficient. Creativity is inefficient. Learning is inefficient. And that’s where the magic is. The goal of technology shouldn’t be to eliminate that inefficiency, but to create more space for it.
So the next time someone tries to sell you an AI-powered solution that promises to automate a deeply human task, be skeptical. Be very skeptical. Ask the hard questions. And don’t be afraid to say, "The emperor has no clothes." Because most of the time, you’ll be right.
’
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.
What experience informs this perspective?
This perspective comes from over a decade of building companies in Silicon Valley, two successful exits (RemoteTeam to Gusto, MovieLaLa to Gfycat), and investing in 200+ startups including Anthropic, OpenAI, and Scale AI. I write about what I've lived.
What's the most common pushback you get on this?
People often push back by citing exceptions or edge cases. And they're usually right that exceptions exist. But building a strategy around exceptions rather than patterns is a losing game for most founders.
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.