There’s a category of mistake that founders in education make, and I’ve been making it. The mistake is to assume that teachers will tell you what they need.
They won’t. Not because they don’t know — they know better than anyone. But because what they need is so much more fundamental than the thing you came to ask about that they don’t think to mention it. You ask about AI tools and they tell you about AI tools. You ask about what’s hard about their job and the conversation changes.
Last month I started a discovery effort with one rule: don’t pitch, don’t validate, just ask and listen. I used a single question as my opening, with variations: what does your week actually look like, and where does it get hard? Then I shut up.
A few patterns emerged that I want to share. Not data — these are qualitative impressions from a small sample. But the patterns were consistent enough that I’d bet on them generalizing.
The thing they brought up that I didn’t ask about
Almost every teacher I talked to spent a meaningful portion of the conversation talking about parents. Not about their students. Not about their administrators. Not about their curriculum. About parents.
The pattern took different forms. I spent two hours last week composing a single email to a parent who was upset about a grade. I have a student whose parents won’t sign the forms for the intervention specialist, and I don’t know how to break through. Half my evenings are taken up by Class Dojo messages that arrive at 9pm and somehow can’t wait until morning.
I had not gone in expecting to hear about parents. I went in to talk about teaching. But the parent communication load is, for most of these teachers, the single largest source of unbillable, unpaid, unvalued work in their week. It’s the cognitive overhead that makes the rest of their work harder.
This isn’t a problem AI is going to solve cleanly. The relationship between a teacher and a parent is delicate in ways that no automation should touch. But the workflow around it — the time spent drafting messages, the cognitive switching cost of moving between curriculum work and emotional labor, the difficulty of finding the right words for a difficult conversation — there might be something to do here. Not by replacing the teacher’s voice, but by reducing the friction around the work they’re already doing.
I’m not building this. But someone should.
The thing I expected them to complain about and they didn’t
I expected to hear a lot about AI cheating. The discourse online is dominated by it. Every think piece on AI in education starts with the assumption that this is the central concern.
It barely came up.
When I raised it directly, teachers had thoughtful things to say. Most of them have caught students using AI; most of them have adjusted their grading. But it’s not the dominant concern in their professional life. It’s a manageable problem — uncomfortable, sometimes frustrating, but manageable.
What’s not manageable is the time crunch. The thing they bring up unprompted, over and over, is time. I don’t have time to differentiate. I don’t have time to grade the way I used to. I don’t have time to plan the lessons I want to plan. I don’t have time to call the parents I should be calling.
This realigned my thinking. The AI conversation in education is being shaped by people who are not in classrooms, and those people are worried about the wrong things. The people in classrooms have a much more boring concern: they need their week back.
Anyone building AI for teachers should start there. Not with cheating. Not with personalization at scale. Not with adaptive learning paths. Just: what would give a teacher two hours back per week, and how can we ship it?
The thing every teacher said that became my new frame
There’s one phrase, in slight variations, that I heard from almost every teacher I talked to. I just want to teach.
It came up after they described some piece of administrative overhead. The data entry. The compliance reporting. The PD requirements. The evaluation prep. The fundraiser coordination. Whatever it was, the teacher would describe it and then sigh and say: I just want to teach.
This phrase is doing a lot of work. It’s saying: the parts of my job that are actually about students learning are the parts I’m best at, the parts that drew me to this profession, the parts I’m willing to stay late for and skip lunch for and take work home for. The parts that aren’t about that are the parts that are eating me alive.
The implication for AI in classrooms, I think, is enormous. The right tool is the one that takes more of the not-teaching work off the teacher’s plate so they can spend more time on the teaching work. The wrong tool is the one that tries to automate the teaching itself.
This sounds obvious when you say it. It is not the design philosophy of most AI products being built for teachers. Most of them try to do some version of the teacher’s job — generate the lesson, grade the work, talk to the student. What teachers actually want is for someone to do the other parts of their job, so they can do the teaching themselves, with their own voice, in their own classroom, with their own students.
I’m noting this here partly because it’s a useful design principle, and partly because it changes my own product thinking in ways that aren’t fully resolved. The temptation to build the impressive thing — the AI that teaches — is strong. The teachers I talked to don’t want that. They want the boring thing that makes the impressive thing they already do, possible.
What I’m doing with this
I’m writing more of these. Slower than I’d like; the conversations take time to set up and longer to process. But the signal is good. There’s a version of this work where I spend the next three months doing nothing but talking to teachers, and at the end of it I have a much clearer picture of what to build than I had at the start.
I’m not going to tell you who I’m talking to or where they teach. They told me things in confidence and with the implicit understanding that I wouldn’t turn them into case studies. But the meta-pattern is fair game, and I’ll keep sharing it as it develops.
If you’re a teacher reading this and any of it sounds wrong — or right, or worth pushing back on — write to me. I read everything. I learn from everything. And the next round of these conversations will be sharper because of it.
Next week: back to the more analytical. We’ll look at what twenty years of education technology adoption can teach us about AI specifically — and why the lessons of the interactive whiteboard era are more relevant than they sound.