We Cut Three of the Four Tools the Week of the Training

The plan had four tools in it. One as the core, one built specifically for the least technical people in the room, and two general-purpose chat apps for everyone else. A week before the session we cut it to one.

Not because the other three were bad. Because a room of teachers with two and a half hours can absorb exactly one thing, and a training that covers four tools produces a faculty that half-remembers four logins and uses none of them. Four tools demoed is worse than one tool learned. It isn't close.

That decision has held up across every training we've run for an organization with no technical staff — a school faculty, an office of consultants, a back office. Here is the rest of what we've learned.

TEACH THE TOOL THAT'S ALREADY INSIDE THE BUILDING

We spent real hours trying to get institutional access to the tools we thought were best, and got stopped by education verification, domain checks, and an account lockout. Meanwhile the organization already had a capable AI assistant sitting inside the productivity suite it pays for. No procurement, no credit card, no new password, no IT ticket. On adoption, "already there" beats "slightly better" every single time. Pick the tool your audience is already licensed for and teach that one properly.

ORGANIZE AROUND THE WORK, NOT THE FEATURES

Most AI training is a feature tour: here's how it handles text, here's images, here's documents, here's research. People leave impressed and change nothing on Monday, because nobody's job is organized by input type. We rebuilt the session around the actual categories of work — making materials, grading, parent and admin communication, and researching a subject you were assigned but never trained in — and taught the capabilities inside those buckets, each one framed as here is what I put in, here is what I expect to come out.

THE REAL CURRICULUM IS PROMPT CRAFT

One teacher told us she'd tried this last year and it failed. She had pointed the tool at the state curriculum framework and asked it to summarize the sixth grade standards. It told her the request was too large. The math framework alone runs 214 pages covering PreK through 12. She concluded the tool wasn't ready.

The tool was fine. The ask was the problem. Point it at a page range. Tell it which document is primary and which ones are reference only, or it will weigh ten files as equals. Teaching that one distinction moves adoption further than any feature demo, so we write a prompt together live, everyone runs it on their own machine, and then people report back what worked and what didn't. That report-back is the most valuable ten minutes of the day.

DRAW THE BOUNDARIES OUT LOUD

The school's own position was that students should not use these tools at all — writing and thinking are the things school is for — while teachers should use them heavily for prep. We built the whole training on top of that line rather than arguing with it.

We added our own lines too. Don't scan a second grader's handwritten worksheet to have it graded; scanning takes longer than grading it by hand. Scope machine grading to multiple choice, typed work, and legible older handwriting, and always instruct it to flag what it can't read instead of guessing. A training that tells people where the tool is a bad idea gets believed about the places where it's a good one.

PRACTICE ON FAKE DATA

Free tiers train on what you put into them, which means real student records, client files, and employee data do not go in. That constraint usually stops a hands-on session dead. So we built a mock data pack instead — a gradebook, a lesson plan tracker, a completed worksheet to grade — with a plantable answer in it so people could see whether the tool found what we knew was there. Practicing on invented data teaches the second lesson for free: you watch it state something confidently and wrongly, in a setting where being wrong costs nothing.

MAKE THEM BRING REAL WORK

The session isn't a demo, it's a build. Everyone arrives with a lesson they are actually going to teach in the first week, and leaves with the materials for it. Fifteen minutes at the top for account setup, because someone always can't log in, and the whole room stalls if you discover that at minute forty.

LOOK ONE DESK OVER FROM YOUR INTENDED AUDIENCE

The people who benefit most are usually not the ones who asked for the training. We scoped this one for teachers. Sitting in the room was the office staff — the people handling scheduling, parent emails, forms, and the recurring reports nobody has time for. Their work is more repetitive, more text-based, and less judgment-bound than the teaching. They may get more out of it than anyone, and they'd have never been the ones to request it. Check the invite list before you finalize it.

NONE OF THIS IS ABOUT THE MODEL

The organizations we work with are not held back by capability; the frontier passed what they need a while ago. They're held back by having four options and no default, a feature tour instead of a workflow, and a first attempt that failed for a reason nobody explained to them.

The measure of a training isn't the energy in the room. It's what's still running in six weeks. One tool, three workflows, on their own files. That's the whole target.