I’m going to say what most EdTech founders are afraid to: your expensive AI grading software is probably useless.
There, I said it.
I’ve seen the demos. I’ve sat through the pitches from hopeful startups as an angel investor. I’ve even tested dozens of these platforms myself. They all promise a revolution in education: instant, unbiased, and scalable grading that will free up countless hours for teachers. It sounds incredible. But the reality is a mess.
Just last year, a founder from a promising EdTech startup pitched me. They had raised a few million dollars and had a slick interface. They showed me how their AI could grade a high school history essay. On the surface, it looked impressive. It checked for grammar, sentence structure, and historical name-dropping. It gave the essay a score of 88/100. The problem? The essay was completely wrong. It confidently stated that the American Revolution was a conflict to preserve the institution of monarchy. The AI saw keywords and good grammar, but it had zero understanding of the actual meaning. It couldn’t spot a fundamental, disqualifying error. I passed on the investment.
The 50% Problem: Why AI Fails Our Students
That experience isn’t a one-off. It’s the norm. The file name for this post has a "50" in it for a reason. From what I've seen, even the best AI graders are barely better than a coin flip when it comes to assessing anything beyond rote memorization. They get the simple stuff right, but they completely fall apart on the nuanced, creative, and critical thinking that we actually want to encourage in students.
These tools are built on a flawed premise. They are optimized for patterns, not for understanding. They turn learning into a game of keyword bingo. Students quickly figure out what the algorithm wants and write for the machine, not for the sake of genuine intellectual exploration. They learn how to hit the rubric points without actually mastering the material. We're training a generation of students to be expert prompt engineers for a dumb machine.
Here’s where they consistently drop the ball:
- They Penalize Creativity: Does the student present a brilliant, out-of-the-box argument that isn’t in the training data? The AI flags it as incorrect. I saw one tool penalize a student’s code submission because they used a more efficient, modern library that the AI hadn’t been trained on. They were punished for being ahead of the curve.
- They Miss Nuance: A history essay isn't a math problem. A literature analysis isn't a database query. These subjects live in the grey areas. AI grading tools are binary; they see right or wrong. They can't appreciate a well-argued but unconventional thesis.
- They Foster a False Sense of Objectivity: We're told the AI removes human bias. That’s a fantasy. It just replaces human bias with machine bias. The AI is trained on a dataset that reflects the biases of its creators. Instead of a teacher’s subjective opinion, you get the cold, unthinking, and often nonsensical judgment of an algorithm.
Stop wasting money on these tools. They're failing our students and burning through cash that could be spent on things that actually work. It's time to stop chasing the fool's gold of fully automated grading.
A Better Way: The AI-Assisted Grading Framework
I'm not an AI pessimist. I’ve invested in over 200 companies, including giants like Anthropic, OpenAI, and Scale AI. I believe AI is the most powerful tool of our time. But it has to be the right tool for the right job. When it comes to grading, the job is not to replace the teacher, but to augment them. To give them superpowers.
After years of advising EdTech companies and seeing what works and what doesn’t, I’ve developed a simple framework. It’s not as sexy as a fully automated grading button, but it actually works. It enhances, not replaces, human expertise.
Step 1: AI for Triage and First Pass
Don't ask the AI to be the judge. Ask it to be the assistant. The first job of an AI in the grading process should be to handle the grunt work. A teacher grading 100 essays has to spend hours on repetitive tasks before they even get to the core intellectual work. That’s where AI can be a massive help.
Use it to:
- Group Similar Submissions: The AI can quickly cluster essays or problem sets that make similar arguments or have similar errors. A teacher can then address the feedback for that entire group at once, saving a ton of time.
- Flag for Plagiarism: This is a perfect task for a machine. It can scan a document against a massive database of sources in seconds. This is a simple, high-value task that frees up a teacher’s time.
- Check for Basic Requirements: Did the student meet the word count? Did they cite the minimum number of sources? Is the code written in the right language? These are simple yes/no questions an AI can answer instantly.
This first pass doesn't assign a grade. It organizes the work for the human expert. It’s about efficiency, not judgment.
Step 2: Human for Nuance and Real Grading
Now that the administrative work is done, the teacher can focus on what they do best: thinking. With the submissions triaged and pre-sorted, the teacher can dive into the actual substance of the work. They can appreciate the creative argument in the history essay. They can see the elegant solution in the coding project. They can understand the subtle literary analysis.
This is the part of the job that can never be automated. It’s where real teaching happens. It’s the human connection and intellectual mentorship that students remember for the rest of their lives. By clearing away the clutter, AI gives teachers more time and energy for this critical work.
Step 3: AI for Scaling Personalized Feedback
Once the teacher has assigned a grade and made the core, high-level comments, AI can come back in to help scale the feedback process. The teacher can write a few keynotes, and then use an AI tool to help flesh them out into more detailed, constructive comments.
For example, a teacher might note: "Good argument, but weak evidence in paragraph 3." An AI assistant could then generate a few variations of more detailed feedback based on that note, such as: "Your thesis is strong and well-articulated. To make your argument even more persuasive, consider strengthening the evidence you provide in the third paragraph. For instance, you could include a specific statistic or a direct quote from the primary source to support your claim."
The teacher remains in full control, selecting and editing the final feedback. This combines the personal insight of a human expert with the scalability of a machine. One of my portfolio companies, which I can't name publicly, implemented this exact model. They saw teachers reduce their grading time by over 10 hours a week, while student satisfaction with the quality and depth of feedback increased by over 30%.
Stop Chasing the Wrong Dream
Let's be honest with ourselves. The dream of a magical AI that can perfectly grade any student work is a long way off, if it's even possible at all. And frankly, I don't think it's a future we should even want. Great teachers are mentors, not grading machines.
The real opportunity in EdTech isn't to build a robot teacher. It's to build a co-pilot. It's to create tools that handle the tedious, repetitive parts of the job and free up our educators to do the deeply human work of inspiring and challenging their students.
So, to the founders building the next AI grading tool: stop chasing the automation fantasy. It’s a dead end. Instead, talk to teachers. Understand their workflow. Find the bottlenecks and the soul-crushing administrative tasks. And then build smart, focused AI tools that solve those specific problems. Build a co-pilot, not an autopilot. That’s a company I would invest in.
Frequently Asked Questions
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.
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.
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.