Eight weeks into my first teaching job, I realized that I had to write personalized report card comments, one paragraph for each of my students in 4th through 9th grade. All 450 of them. Two weeks sounds like plenty of time. But I am a procrastinating writer. I agonize over every word. The best way to get me to clean my house is to give me a writing task.
I don't remember how I did it. I'm sure I asked for every extension available. But I remember the feeling. The thing I was supposed to be doing, telling each student something true and specific about their learning, was mentally and physically impossible. Since that first experience with report cards, I have been working on easier ways to get this done without compromising the authenticity and utility of each comment.
Over the years I built comment banks, mail merges, batch grading routines. I even learned programming languages just to automate parts of the task. I got faster. But faster isn't the same as better. Every one of those tricks was just a more efficient way of coping with the same impossible math: the feedback that actually matters to students and parents, the specific, personal, timely kind, is the feedback that takes the most time to write. And every hour I spent on comments was an hour I wasn't reflecting and working with students.
Something Changed
Something changed in January 2026. I started using Claude Code.
I'd been messing around with AI tools before the release of ChatGPT. I even built and deployed a tool that uses rubrics and submission comments to draft report cards. But this year the technology got good enough that I stopped asking "how do I get through this grading pile faster" and started asking something different: "does this task actually need me?"
I literally went through my workday task by task and tried to automate everything I could. I trained an AI on 462 of my actual comments from past semesters. Not a generic "teacher voice." My voice. My patterns, my phrasing, the way I balance honesty with encouragement. Then I built a system around it. It pulls student submissions, scores them against the rubric, drafts feedback that sounds like me, and shows me everything before a single grade gets posted. I review every piece of feedback. I change what needs changing. But I'm editing now, not writing from scratch. That's a different job.
Twenty Tools in Three Months
Grading was just the beginning. Over the past three months I've built over twenty tools. AI Skills and Agents that handle lesson planning, content creation, competency tracking, and course audits, all wired directly into our school's learning management system through API connections. Every tool respects student privacy. Anonymization is built into the data layer, and sensitive documents like IEPs and 504 plans never touch the AI.
I built a tool that turned a chapter of a Spanish novel into 31 quiz questions in minutes. I built another that generated an honest transition report when a colleague left mid-year, showing the substitute exactly what had been taught, what hadn't been graded, and where each student stood. I built a gamified grading interface that turns the feedback loop into an 8-bit video game. I started recording my classes (with student consent) and feeding the transcripts to three AI-simulated educational philosophers (Montessori, Dewey, and Papert) who review every lesson and tell me what worked and what didn't. Then the system updates my next lesson plan and Canvas content on its own.
Maybe the tool I'm most proud of is one that builds a narrative portrait of each student's growth across all their courses. It pulls every rubric assessment, every teacher comment, and actual excerpts of student work, then assembles a "Portrait of a Graduate" that tells the story of who that student is becoming. No percentages. No point values. Just rubric level names, real teacher quotes, and the student's own writing. The kind of document I always wished I had time to create for every kid.
Over twenty tools. Three months. One procrastinating writer who got tired of staring down report card comments by hand. The tools aren't the point, though. What I'm able to do with the time is.
The Work That No AI Can Do
When I wasn't spending my time grading, I built interactive explorable explanations inspired by Bret Victor, where students discover how stochastic gradient descent works by dragging sliders and watching the math respond in real time. When I wasn't manually assembling and posting Canvas modules, I was rethinking what my courses should actually feel like. When I wasn't buried in the administrative side of teaching, I was doing the work that no AI can do: designing experiences that make hard concepts click, having real conversations with students, being the kind of teacher I got into this for.
The tools won't replace me. They cleared away the busywork so I could do the job I actually signed up for.
What's Next
Over the next several weeks, I'm going to share everything I built. How it works, what failed, what surprised me. Not because the technology is the story. But because it points at a question every school is going to have to answer soon: if AI can handle the administrative and algorithmic parts of teaching, what should teachers actually be doing with their time?
I have an answer. It starts with what my school calls its nine transdisciplinary skills. The Human Superpowers.