How Do You Teach a Team to Guide AI, Not Just Use It?

Quick Answer

 

How do you teach a team to guide AI, not just use it? A former English teacher turned COO on modeling, practice, feedback and where judgment stays human.

How Do You Teach a Team to Guide AI, Not Just Use It?

Every English teacher has graded the essay that sounds finished and says nothing. The sentences are clean, the transitions are polite, and somewhere around page two you realize the writer never made an argument. I read a lot of those over 22 years in the classroom, and AI can now produce them at a pace no sophomore ever managed.

The habit is spreading faster than the training. Three in four knowledge workers were already using AI at work in 2024, according to Microsoft and LinkedIn's 2024 Work Trend Index, yet only 39% of those users said their company had given them any AI training. As COO of Hollywood Branded, I build AI workflows across Microsoft 365, PandaDoc and Claude, and the part that has to be taught is guiding the tool: setting the task, judging the output and owning the result. In this article, I explain how I approach teaching a team to guide AI, using one of the oldest lesson plans in education.

Quick answer: Teaching a team to guide AI works best with the classroom model of gradual release: the leader models the full workflow, the team practices it together on real work, then individuals run it alone with feedback. Ian Drummond, COO of Hollywood Branded, says the goal is judgment: people who can direct AI, check its work and own the final result.


The Lesson Plan Is Older Than the Software

The mistake most companies make is skipping straight to the last step. They hand people a license, maybe a list of sample prompts, and expect good work to follow. That's like handing a class a stack of novels and calling it a literature course.

Teachers solved this problem a long time ago. In 1983, P. David Pearson and Margaret Gallagher described what's now called the gradual release of responsibility, which most teachers know by its shorthand: I do, we do, you do. The teacher models the skill, the class practices it with support, and then students work on their own while the teacher watches and coaches. It's the backbone of how a lot of English teachers teach essay writing, and it maps almost perfectly onto AI adoption.

The sequence works for a simple reason. People who watch an expert first learn what good looks like, people who practice with support learn where they go wrong, and people who finally work alone have already made their early mistakes somewhere safe. The tools will change every few months, and the method holds up fine without an update.


What Does Modeling Look Like When the Tool Is AI?

Modeling means showing the whole loop out loud: the instructions you gave, the first draft that came back, and every edit you made, with the reasoning attached. The prompt is the least interesting part. The cuts are the lesson.

When I model a workflow, whether it's summarizing a long vendor document, prepping a contract checklist or drafting a first pass at a policy, the most useful thing I can show is where the output was wrong, vague or too confident. AI is very good at sounding finished, a skill it shares with a lot of sophomores. A team needs to see someone they trust say "this sentence isn't true" and "this one promises something we never agreed to."

That moment teaches more than any prompt library, because it gives people permission to doubt the machine and shows them what judgment looks like in practice.


How Do You Move From Practice to Independence?

The "we do" stage is where most of the learning happens. Pair on real work, not exercises: an actual summary, an actual draft, an actual spreadsheet that has to be right. Review the output side by side and talk through what to keep, what to cut and what to check against the source. Feedback should be specific and quick, the same way it is when you hand back an essay with notes in the margin instead of a single grade at the top.

Then let go, with a check. The "you do" stage works when people have a short list of things to verify every time: facts and figures traced to a source, names and dates confirmed, nothing promised that hasn't been approved. I'd put one rule in front of any team: every AI-assisted document has a human owner, and that person's judgment is the final edit.

Some calls stay human no matter how good the tools get. Relationships, negotiation positions, creative judgment and anything with a signature on it belong to people. Anyone who has redlined an MSA knows a single word like "exclusive" can carry an entire deal. AI can draft the email. It shouldn't decide what the email asks for.


Why Does This Matter for a Creative Agency?

When a workflow lives in one person's head, it walks out the door with them at 6 p.m. Once it has been modeled, practiced and checked, it becomes a system, and systems are what let a small team do consistent work at scale. That matters at Hollywood Branded, where a core team of 30 works alongside the [HB]Network of specialists and former studio executives, and the work spans brand integrations, licensing, celebrity and influencer partnerships, events and more.

It's also why I made AI use mandatory in our hiring assessments. The test is how a candidate directs a tool, questions its answer and decides what to keep, because that's the same judgment we teach once someone is on the team. A clever prompt earns about as much credit as a nice font on a term paper.

For brands, the benefit is practical. When the first draft of a recap, a summary or a contract checklist takes less time, more of the team's hours go to strategy, relationships and creative thinking, which is the work clients actually hire an agency to do.


Start With One Workflow

You don't need a company-wide AI program to begin. Pick one task your team does every week, model it from start to finish, run it together twice on real work, and then hand it off with a short checklist of what must be verified.

Do that well once, and you'll have a template for the next workflow and a team that knows the difference between using a tool and being responsible for what it produces.

New call-to-action


Frequently Asked Questions

Q. What is the best way to train employees to use AI?

A. A reliable approach borrows the gradual release model from education: a leader models the full workflow, the team practices it together on real tasks, and individuals then work independently with feedback and a verification checklist. Training on real work builds judgment faster than prompt lists or generic demos.


Q. What is the difference between using AI and guiding AI?

A. Using AI means asking a tool for output and accepting what comes back. Guiding AI means setting a clear task, checking the output against sources, editing what is wrong or overstated, and owning the final result. The second skill is what makes AI safe and useful in professional work.


Q. Which business decisions should not be handed to AI?

A. Relationship decisions, negotiation positions, creative judgment, approvals and anything a person will sign should stay with people. AI can draft, summarize and organize information for those decisions, but a named human should own the call and verify the facts behind it.


Eager To Learn More?

Here are a few more reads on strategy, partners and staying current:

Want to stay in the know with all things pop culture? Look no further than our Hot in Hollywood newsletter! Each week, we compile a list of the most talked-about moments in the entertainment industry, all for you to enjoy!

New call-to-action

 

Table Of Contents

    WHAT'S HOT IN HOLLYWOOD

    Want to keep your pulse on what Hollywood has to offer? Sign up below to get on our HOT IN HOLLYWOOD partnership updates!