A few weeks ago, I watched a colleague spend forty-five minutes building a quiz. She had the learning objectives written down, she knew which standards she was targeting, and she’d already annotated the readings. She knew exactly what she wanted. She just had to do the thing — type the questions, write the distractors, format the document, upload it. Forty-five minutes of her Saturday afternoon for something that would take her students twelve minutes to complete.
I didn’t say anything at the time. But I kept thinking about it.
Because that forty-five minutes isn’t a tools problem. It’s a workflow problem. And in 2026, that distinction is starting to matter a lot.
The Shift Nobody’s Talking About Loudly Enough
Here’s what’s actually changing right now: AI is moving from something you prompt to something that acts.
For the past couple of years, the conversation around AI in education has been about tools — ChatGPT for lesson ideas, MagicSchool for rubrics, Brisk for feedback in Google Docs. You type something in, you get something back. You’re still the one doing the driving.
That’s changing fast.
What’s emerging now — and what the education world hasn’t fully reckoned with yet — is agentic AI: systems that don’t just respond to a single question, but complete sequences of tasks on your behalf. Instead of asking AI to write one quiz question, you describe what you need and the agent plans the assessment, writes the questions, calibrates the difficulty, formats the document, and drops it into your LMS. While you grade papers.
Canvas recently launched an AI teaching agent — called IgniteAI — that can handle rubric generation, content alignment, and discussion review autonomously. MagicSchool’s new “Knowledge” feature lets administrators upload curriculum documents once and have them apply across every tool in the platform. Brisk now integrates directly into Google Docs and automatically aligns to your school’s pacing and standards — without you having to tell it anything.
These aren’t chatbots. These are co-workers.
Why This Should Excite You (and Also Make You Thoughtful)
I’ll be honest: the first time I realized how much had shifted, my reaction was equal parts “this is incredible” and “wait, slow down.”
The incredible part: a recent survey found that teachers who use AI effectively save between five and ten hours per week. That’s a unit plan. That’s sixty student feedback drafts. That’s a week’s worth of Sunday anxiety about Monday morning. The promise of agentic AI is that those time savings compound — because the agent isn’t just doing one thing, it’s doing a chain of things that used to each require your attention.
The thoughtful part: with great automation comes the risk of quiet drift. When AI starts completing tasks on your behalf — not just helping you complete them — the question of whose judgment is in the work gets more complicated. A rubric generated autonomously and applied to student work without your review isn’t really your rubric. Feedback drafted by an agent and sent directly to a student’s document without your read-through isn’t really your feedback.
The research is pretty clear on this: hybrid human-AI workflows produce better outcomes than fully autonomous ones. The agent can draft, organize, and generate. But the teacher has to be in the loop — not just technically, but meaningfully.
What This Looks Like In Practice
Let me give you a concrete example of how I’m actually using this.
When I’m preparing feedback for a round of AP World History DBQs, I don’t start from scratch. I describe the assignment, paste in the rubric, summarize what patterns I’m seeing across the class, and let AI generate a first-draft comment for each student. Then I read every single one. I revise them. I add the specific detail that only I know — the fact that this student has been working on thesis clarity for three weeks, or that this one got flustered during Socratic seminar and needs encouragement, not critique.
The agent gets me 70% of the way there in a fraction of the time. I do the 30% that actually makes the feedback worth receiving.
That’s the workflow I want teachers to understand. It’s not AI replacing your judgment. It’s AI doing the scaffolding so you can spend your energy on the part that requires a human.
The Equity Question We Can’t Skip
Here’s the thing I don’t hear enough people saying: agentic AI is not neutral.
When AI systems are trained on existing data — existing lesson plans, existing assessments, existing classroom norms — they encode the biases of whoever created that data. A “personalized learning path” that sounds objective might systematically steer certain students toward remediation. An automated rubric that applies consistently across a class might consistently misread the writing patterns of students whose first language isn’t English.
I teach students from dozens of countries. Many of them write in ways that are perfectly sophisticated but don’t sound like what a model trained on American academic prose expects. If I outsource my assessment to an agent without interrogating how it evaluates “good writing,” I’m not being efficient. I’m being negligent.
Agentic AI makes this more urgent, not less — because the more autonomous the system, the more decisions get made without human review. And the more decisions get made without human review, the more bias can quietly accumulate in student outcomes.
This isn’t an argument against agentic AI. It’s an argument for staying in the loop. For auditing the work the agent does, not just accepting it.
The Honest Bottom Line
I think we’re at an inflection point that doesn’t come around often in education. The tools are genuinely starting to change what’s possible for teachers — not just in theory, but in daily practice.
But the schools and teachers that will actually benefit are the ones who engage with this thoughtfully. Who understand what the agent is doing and why. Who stay curious about the edges — what it gets wrong, who it might disadvantage, what it can’t see that you can.
My colleague who spent forty-five minutes building that quiz? She could get that time back. And I want her to. I just want her to read the quiz before it goes out.
The Takeaway for Teachers
Try this this week: pick one repetitive planning task — a quiz, a feedback template, a discussion prompt set — and let an AI tool build the first draft for you. Don’t start from scratch. Describe what you need clearly (learning objective, grade level, specific skill or standard), let the tool generate it, and then spend your time revising rather than creating.
That revision step is non-negotiable. Read what it made. Fix what’s off. Add what only you know. But notice how much faster you got to the finish line — and what you could do with that time instead.
That’s the shift. Start practicing it now, before the agents get even more capable.
David Jacobson teaches AP World History at Shanghai American School. He writes about AI, education, and the messy intersection of the two at shouldiuse.ai. Follow him on LinkedIn and Substack.
