Steal My AI Workflow: Five Areas I’m Using AI For Professionally

Most AI advice for educators is still stuck at the prompt level.

You have seen the posts. "50 prompts every teacher needs." "The ultimate prompt library for principals." I have written some version of that myself (sorry!), and I do not think it was wrong so much as it was small. A prompt gets you a better sentence, BUT a workflow gets you back a Tuesday afternoon.

The shift I have made over the last two years has almost nothing to do with finding better words to type into a box. It had to do with figuring out which parts of my job are actually repeatable processes wearing a disguise, and then handing those processes off in a way that holds up when I am not watching.

Here are the five places that shift has stuck. Steal any (or all) of them.

One more thing. I walk folks step-by-step through these systems (and much more) in my AI-Ready School Leadership Certification Program on Maven. New cohort starts this Fall. In just a month’s time, you’ll be fully up to speed like all the folks are sharing in their reviews of the program here.

1. Cowork: Handing Off The Whole Job, Not The Sentence

The mental model most people carry is chat. You ask, it answers, you copy, you paste, you fix. That loop is fine for a paragraph but actually terrible for a project.

Cowork changes the unit of work. Instead of asking for an email, I hand over the task (with some caveats) Cowork allows me to say “here is the folder of session notes from the last six weeks of practices", here is the template we use, build the parent facing season overview and flag anything that contradicts what I told families in September”.

I just did that for the lacrosse team I coach.

That is a task with inputs, a standard, and a check. It runs across multiple steps and multiple files, and I come back to something I can edit rather than something I have to build.

For a school leader, the equivalent is, “Here are the 12 department improvement plans; pull every stated goal, group them by our strategic priorities, and show me which priority nobody claimed”.

That is a job that eats a full day of a curriculum director's week and produces a document nobody reads. Now it takes 20 minutes and produces a document you argue about, which is a much better use of a leadership meeting.

Steal this: Pick the one recurring task in your role that involves reading many documents and producing one. That is your first handoff to Claude Cowork. Haven’t used Cowork before, or don’t use Claude? You can get it here for free.

2. Deep Research: The Literature Review That Used To Take A Week

I am writing a book right now that leans hard on learning science, and I have a rule about citations. If I cannot link it, I do not claim it. That rule used to be REALLY expensive. It meant hours of chasing a study back to the source, finding out the effect size was smaller than the headline, and quietly cutting a paragraph I liked.

Deep Research compressed that entire timeline. I ask a real question. Something like: what does the evidence actually say about the relationship between student engagement measures and achievement outcomes, including the studies that failed to replicate. Then I get a sourced report I can interrogate.

The critical part is what happens next, and this is where I see people misuse it. I read the sources. All of them. The research pass is the reading list assembled by something faster than me. I have caught confident summaries of retracted work more than once. The tool is fast (not infallible), and the difference matters more in our field than almost any other.

For educators, this is your curriculum vetting tool. Before you adopt the program, before you sit through the vendor pitch, ask what the independent evidence base looks like and who funded the studies. You will walk into that meeting as the most prepared person in the room.

Steal this: Next time you are about to accept a claim in a professional development session, research it that night (or that day). Bring the nuance back the following week.

3. Skills: Teaching It My Repeatable Processes

This is the one almost nobody in education is using yet, and it is the one I would push hardest.

A Skill is a set of instructions you write once that the AI follows every time a particular kind of work shows up. It is an actual process you encode. You describe how the work should be done, what a good version looks like, what to never do, and then that standard applies every time without you rebuilding it from scratch.

Start with email, because email is where the week actually goes.

The first Skill worth building is triage. Mine reads what came in overnight and sorts it into four buckets…

Bucket 1: This needs a real reply from me.

Bucket 2: This needs a two line acknowledgment.

Bucket 3: This needs to become a calendar item.

Bucket 4: This needs nothing.

The Skill surfaces the first bucket and drafts the second. What used to be forty minutes of deciding is now ten minutes of deciding plus a few edits.

For a school leader, the buckets are different and the logic is the same. Family concern that needs a response today. Staff question somebody else on the leadership team should own. District compliance request with a deadline buried in paragraph four.

The Skill knows the difference because you told it once, and because you also told it your escalation rule where anything mentioning a student safety concern gets flagged to the top regardless of who sent it and never gets an automated draft.

The second email Skill is the follow-up loop. It tracks what you have not answered after two days and reminds you. That sounds trivial until you count how many parent relationships have quietly soured because a message got buried under 200 others during a hard week.

Beyond the inbox, look for anything you write more than ten times a year with a stable structure and a changing subject.

One rule I would not bend on. Nothing sends without a human reading it, and nothing involving a specific student's confidential information goes into a system your district has not cleared. This is mandatory at all times.

Steal this: Open your sent folder. Find the message you have written thirty versions of this year. That is your first Skill.

4. Building The Tool Instead Of Looking For One

I spent years searching for software that did the specific thing I needed and then settling for software that did eighty percent of it.

That habit is obsolete and I am still catching myself doing it.

Nowhere is this more obvious than data dashboards. Almost every district I work with has paid real money for a data platform, and almost none of them have the view they actually want. You get what the vendor decided mattered, which is usually a proficiency percentage by grade level, which is usually the least actionable number in the building. Everybody nods at it in the September meeting. Nobody changes anything on Tuesday because of it.

Here is what changed. Your student information system exports a spreadsheet. So does your assessment platform. So does your attendance system. The moment you can export it, you can describe the view you have always wanted and have a working version of it that same afternoon.

The first one I built with a Principal took about twenty minutes. She did not want the chronic absenteeism report, because by the time a student appears on that report the pattern is already set. She wanted the students sitting between eighty-five and ninety percent attendance. The ones about to cross the line. That is the intervention window, and no platform she had access to would show it to her, because it is not a number anybody has to report to the state.

The second was a single screen that put attendance, behavior referrals, and course grades side by side for each student, because the early warning signal has never been any one of those. Three separate systems will never tell you that a kid slipped in all three at once during the same two weeks in November.

For teachers, the version I would build first is mastery by standard by student. A grid that says these six kids missed this specific standard, sorted so you can form tomorrow's small group in ninety seconds instead of forty minutes with a highlighter. The second one is a contact log that shows you which families you have reached out to, when, and what the ratio of positive to concerned messages looks like. That ratio is uncomfortable the first time you see it, which is exactly why it is worth seeing.

Two ground rules before anyone runs off and does this. Use the data you already have legitimate access to, and keep it inside whatever environment your district has cleared. The point is to see your own data more clearly, not to move student records somewhere new.

None of these are impressive pieces of engineering. They are small, ugly, and exactly right, which is a category of tool that has never been economically viable to build before, but it is viable now.

The real unlock is how you iterate. When the dashboard is wrong, and it will be, you say what is wrong and it changes in two minutes. Compare that to a feature request that goes into a vendor queue and comes back in a version release eighteen months from now, if it comes back at all.

Steal this: Export one file you already have access to. Describe the single view you have always wished someone would build from it. Make sure you De-Identify when you need to!

5. The Pressure Test Before The Decision

The last one is the least flashy and the one I would give up last.

Before I send the proposal, launch the program, or walk into the hard conversation, I ask for the strongest case against it. This is a Pre-Mortem. I’ll aks it to argue that this initiative fails, and tell me what the first sign of failure looks like.

I did this with a program pitch I spent a month writing and got back three objections I had not considered, two of which came up almost verbatim in the actual meeting. I did it before a difficult conversation with a coach and had it play the other side, which was uncomfortable in a useful way.

Leadership is full of decisions that get one rehearsal, and that rehearsal is usually the real thing. This is the closest thing I have found to a second attempt.

It is very easy to use this tool as ONLY a mirror. Ask it whether your idea is good and it will find reasons. Ask it to kill your idea and you learn something. The prompt that starts with "what am I missing" is worth a hundred that start with "is this good."

Steal this: Before your next consequential decision, ask for the premortem. Six months out, this failed. Why?

The Through Line

Look back at those five and notice that I’m still doing the thinking. Deep Research assembles the reading, I still read. The Skill drafts the coaching email, I still decide what a teacher needs to hear. The pressure test surfaces the objection, I still choose whether it changes my mind.

More of my week is judgment and discernment, and less of it is assembly and organizing. That is literally the whole promise, and it is a real one.

Every one of these workflows gives me back capacity, and capacity is not neutral. It goes somewhere. If it goes into producing more polished materials that students still have no say in, we have automated Learning 2.0 and made it shinier. That would be a genuine loss dressed up as a win for many folks.

The reason I care about any of this is that the time it returns is the time we do not currently have for the work that actually matters.

I hope you steal the workflows and then be deliberate about where the hours (and mental energy) goes.