The actual teacher week is 56 hours, give or take. About 25 are with students. About 7 are mandatory meetings, prep periods, lunch duty, and bus duty. The remaining 24 are work — mostly at home, often late.
Of those 24 hours, here's where they go for the average elementary or middle-school teacher:
- 5-6 hours: lesson planning and prep
- 4-5 hours: grading and feedback
- 3-4 hours: IEP/504 documentation (for sped or co-teaching) or assessment-data analysis (for others)
- 2-3 hours: family communications
- 2-3 hours: admin (forms, surveys, professional learning prerequisites)
- 4-5 hours: the buffer — when one of the above takes twice as long, this is where it comes from
AI can reasonably take a real bite out of three of those buckets. Let's go through them.
Lesson planning (1-2 hours back)
AI shouldn't write the lesson. It should differentiate the lesson the teacher writes. Tier the same plan for three reading levels. Generate the modified version for the EL learners. Produce the extension activity for the kids who'll finish in twelve minutes.
The teacher's IP is the lesson. The AI's value is the variations. This boundary keeps teachers in the work; cross it and you get resistance.
IEP and 504 documentation (2-3 hours back)
This is where the most time can be reclaimed and where the change matters most. A standard 12-hour IEP is mostly compliance language and structured data — both of which AI handles well. The 4 hours that should stay human are present levels, goal-setting, and the meeting itself.
Anything that gets that 12 down to 5 is a teacher-retention win. The senior special-ed teachers we've talked to don't want to do less of the work; they want to spend more of their time on the parts of the work that matter. AI lets them.
Family communications (1-2 hours back)
The weekly classroom update. The progress notes. The conference prep. The translated newsletter to multilingual families. Each of these is the teacher writing roughly the same thing slightly differently for 28 different recipients.
AI handles the "slightly differently" part. The teacher provides the judgment — three things to mention about each kid — and the AI produces 28 personalized notes in 15 minutes.
The best AI in a classroom doesn't make the teacher more efficient. It makes the teacher more themselves. Less time on forms, more time on relationships.
What AI shouldn't touch
The lesson itself. The relationship. The judgment about a specific kid. The conversation with a worried parent.
These are not workflow problems; they're the work. Anyone selling AI that does these things is selling something to a district leader who's never taught.
What this looks like for a principal
If you're a school leader thinking about adopting AI:
- Don't roll out tools to teachers. Ship workflows that teachers ask for. Survey first.
- Pick the three most painful: typically differentiation, IEPs, and family comms. Skip the rest.
- Pilot with willing teachers who already do these well. The senior teacher who's been writing fast IEPs for ten years is your best champion.
- Measure the right thing. Don't measure tool adoption. Measure hours given back, teacher-satisfaction scores, and turnover.