What I Actually Use

I use Claude, made by Anthropic. What changed about my work isn’t that I found the right app. It’s that I use Claude in three different forms, and each one does a different kind of job. Most teachers only know the first form and assume that’s the whole thing. That’s the form that’s changed my week the least.

Here are all three, and what each one is for.

Chat

The chat window

This is the version everyone pictures. You open a tab, you type a question, it answers. Most people use it like a faster search engine, which is the quickest way to get the least out of it.

I use it as a thinking partner. When I’m working out how to frame a unit on decolonization, or trying to figure out why a lesson fell flat, I talk it through with Claude the way I’d talk it through with a colleague who has time and never gets bored of my questions. I argue with it. I ask it to push back. I paste in a paragraph I’ve written and ask where the logic is thin.

The shift that matters is that I’m not asking it for answers. I’m using it to push my own thinking further than I’d push it alone. Students don’t get the difference until they’ve felt it. Neither did I, at first.

This is the front door. It’s useful, and it’s the least interesting room in the house.

Desktop

Cowork: Claude on my desktop

Cowork is Claude living on my computer, able to work with my actual files. The difference between the chat window and Cowork is the difference between asking someone for directions and handing them the wheel.

In the chat window, Claude tells me how to build the thing. In Cowork, it builds the thing. I point it at a folder of messy notes and it turns them into a clean lesson plan in a Word doc. I describe a quiz and it produces one, formatted, ready to print. I ask it to read through a stack of articles and pull the three that fit my Industrial Revolution unit.

It can also work while I’m not watching. I hand it something at the start of a free block, go teach, and come back to a draft waiting for me.

For a teacher, this is where AI stops being a clever toy and starts giving you time back. Not by thinking for you. By doing the production work that was eating your evenings: the formatting, the drafting, the turning of an idea into a document that never needed your judgment in the first place.

Developer tool

Claude Code: the developer tool I’m not a developer for

This is the one that sounds like it isn’t for teachers. I’m a history teacher. I didn’t write code before last year. I still don’t really think of myself as someone who does.

Claude Code is built for software developers, but what it actually does is let you describe a problem in plain English and build a tool that solves it. So I did. I built a grading system that takes a class set of essays, runs my own first-pass notes through a pipeline of graders and critics, and hands me back polished feedback in my voice, calibrated to the AP rubric. I built dozens of smaller tools: one that pulls up today’s lesson, one that drafts my email replies for me to review, one that plans a unit from scratch and critiques its own work.

None of these existed when I started. I described what I needed, and the tools were good enough that a non-technical person could build them. I wrote that whole setup up separately, if you want the details: What My Claude Setup Looks Like.

What this looks like in an actual week

The sub day. I wake up to a sick kid at home. Instead of leaving my classes a video and a worksheet, I have Cowork build a real plan a substitute who’s never heard of the Mughal Empire can actually run, and the day still moves the unit forward.

Feedback. I read every essay myself and make my own call. What I’m reviewing is a stress-tested first draft instead of a blank page, so the time goes to talking with students about their writing instead of typing the same comment for the fortieth time.

The note home. The quiet kid who finally spoke up in the Cold War debate. That note never used to get sent, because it was never the most urgent thing on the list. When drafting it costs almost nothing, it actually goes out.

Two courses at once. I teach AP World History and AP U.S. History in the same week, each running on its own timeline toward May. Re-pitching one idea for two different courses is exactly the kind of work worth handing off.

Where it stops

None of this knows my students. It doesn’t know that the kid in the back row writes circles around everyone when the topic is something he picked himself. It doesn’t know which class needs the debate and which one needs ten quiet minutes. That read is mine, and no tool is close to taking it.

That’s the whole point, and it’s the argument of my book, The AI Doesn’t Know Your Students. The tools handle the work that was never the teaching. What’s left is the teaching.

If you want to go further with any of this, that’s what ourai.club is for.

How the pieces fit together

People sometimes ask if any of this is one tool, or one clever prompt. It isn’t. It’s a handful of purpose-built processes, each one built for a task I do often enough that a one-off prompt stopped being worth it. Here’s the shape of it.

Repeatable pipelines, not one-off prompts. I’ve built a small number of processes for the recurring work: one drafts feedback on a stack of essays against my rubric, one turns a unit outline into lesson plans and discussion questions, one clears a backlog of email into draft replies, one researches a property before I make an offer, one drafts a recommendation letter from what I actually know about a student. Each one runs the same shape underneath: draft, check its own work, revise, then stop. Nothing sends or publishes itself.

A filter that catches AI on the page. Before any of that reaches a student, a parent, or a reader, it runs through a list of tics I’ve built by hand from real drafts: the hedge words, the false enthusiasm, the sentence rhythm that reads like nobody in particular wrote it. I keep adding to that list.

A memory that carries over. A new conversation doesn’t start from zero. There’s a standing layer of notes on my courses, my current unit, decisions I’ve already made, so I’m not re-explaining context every time I open a new window.

A queue, not a firehose. Drafted work lands somewhere I have to look at it before it goes anywhere real. That’s the whole difference between a tool that does your work and a tool that hands you a draft of it.

The right tool for the job, not the biggest one. Formatting and quick lookups get a fast, cheap model. Anything with real judgment in it, or anything going out under my name, gets the careful one. That’s a cost decision as much as a quality one.

A way to keep learning. I keep a running, cross-referenced collection of what I’m reading about AI in education, so when something changes in the field, I can find what I already know instead of starting over.

None of this is exotic. It’s closer to how a well-run department runs: a process for the recurring work, a style guide, someone checking it before it goes out the door. I just built mine with Claude instead of a staff.