Before the Seed Prompt: Why the First 10 Minutes of an AI Workshop Are Where You Lose the Room

I ran a private dry-run of an advanced AI training session this week with three senior leaders before rolling it out company-wide. The most useful feedback wasn’t about the AI.

(The training described here is one I’m running at my employer — Jacobs Entertainment / Lodge Casino — but the broader pattern is generic enough to apply anywhere you’re rolling out an AI workshop to non-technical staff.)

setup

The session sets up non-technical employees with their own personal AI agent. Hands-on workshop format. By the end of about 60 minutes, each attendee walks out with a working agent that knows their projects, their people, their calendar. A “second brain” they can talk to and that talks back with their context loaded.

It’s the follow-up to an intro AI session I’d already run with teams. The intro was about co-pilot-style assistance: ask the chatbot a question, get an answer. The advanced version goes one layer deeper — installing the agent infrastructure, dropping in a seed prompt that loads the user’s context, then teaching the agent its first few skills.

Solid material. Real value at the end. The kind of session that’s actually rewarding to give.

So I did the dry-run with three senior leaders before pulling the trigger on a full rollout. Wanted to catch structural issues before the real audience.

the first 10 minutes lost the room

Not because the content was wrong.

Because the install phase elevated everyone’s anxiety.

One leader said it out loud, almost in passing — “for some reason that elevates my anxiety level.” He was watching the slide that said “install this plugin, then this connector, then this extension.” Nothing technically hard. Just a lot of steps, presented as instructions to follow on his own laptop, in real time, in front of his peers.

Another participant’s machine misbehaved on one of the installs. He fell behind. The whole class shifted into a “we’re waiting for him” energy. The presenter (me) tried to keep going. The participant kept troubleshooting. The other two attendees kept glancing over to see whether they should help or whether they were missing something on their own machines.

Afterward he told me, plainly: “It just turned into a working install session.”

That sentence stuck with me harder than anything else from the dry-run. Because he was right. The first 15 minutes of the session — even though the slides were good, even though the visual design was strong, even though the agent setup itself was clean — were functionally an IT support call disguised as a class.

the inflection point

The session didn’t actually start until I dropped in the seed prompt.

The seed prompt is a long block of text that loads the user’s context into the agent — their projects, their team, their meeting cadence, their company. The first time it gets pasted in, the agent comes alive. It starts asking questions back. It starts knowing things.

The same participant who’d fallen behind on installs — the same one who’d said it was just an install session — looked up after the seed prompt landed and said: “Once the seed prompt went in, it started to make more sense for me.”

That’s the inflection point of the whole session. Before the seed prompt: an install workshop. After the seed prompt: a real agent.

And that’s exactly the thing the attendees didn’t know at minute one. They had no idea the install phase was the cost they were paying for the magic moment 15 minutes later. So they sat in the install phase reading it as if it was the whole class.

what one of the senior leaders suggested

The same leader who’d named the anxiety came back at the end of the dry-run with a brainstorm:

“I wonder if we could take the first 10 minutes, right? And go, okay, maybe we bring one more IT guy into the class. That IT person signs on as someone who goes through making sure that everything’s installed for each attendee — Chrome plugin, the workspace, the connectors. And then you start your conversation with that work.”

It’s a small idea. It changes the session shape entirely.

Right now, in the dry-run version, the presenter is doing both jobs: explaining the content AND walking through the install steps on screen. The attendees are supposed to be following along on their own machines, doing the installs themselves, while also tracking the explanation.

That’s asking a non-technical attendee to do two things at once. One of them is high-anxiety (the keyboard work, the “did I click the right button” loop). One of them is conceptual (the explanation of what the agent is, what it’ll do, why it matters). When you stack high-anxiety with conceptual, the anxiety wins. The conceptual content washes out.

The fix: split the roles. The presenter explains. A second IT person, in the room (physically or on the call), handles the actual installs for each attendee in parallel. The attendees don’t have to touch their keyboards in the first 10 minutes. They get to listen. The anxiety drops out.

The Pit Crew model. I started calling it that internally because the metaphor lands — the driver focuses on driving, the pit crew gets the car ready. Same dynamic.

the opener has to promise the destination

The other change I’m making is simpler. Just a slide.

The current opening slide is something like “Welcome to AI Training — Agenda.” That’s the kind of opener that elevates anxiety because it doesn’t tell anyone what they’re getting. It just announces that they’re doing a class.

The new opening slide tells them the destination:

By the end of this hour, you’ll have an AI agent that remembers.

Not a chatbot that forgets you. An agent that knows your projects, your people, your routines — and can act on its own while you sleep.

Then four small icon-and-line items underneath:

• Remembers. Your projects, your people, your decisions — across sessions, forever.
• Acts on schedule. Morning brief in your inbox. Weekly rollup on Fridays. Done while you sleep.
• Grounded in YOU. Reads your daily notes, your calendar, your team’s chat — not a generic internet bot.
• You stay in the loop. It surfaces, you decide. Your judgment, just faster.

The point of that slide isn’t to oversell. It’s to give the attendees a destination they can hold in their head while they’re sitting through the install phase. So when their machine misbehaves on slide 6, they don’t read it as “this class is going badly.” They read it as “we’re getting through the install part before the part the deck promised at the start.” That re-anchors the experience.

And in the speaker notes for that opening slide, I include this:

“We’re going to install some stuff in the first 10 minutes, and [pit crew name] is going to be doing that with you — you don’t have to keep up with the keyboard work. Just listen and watch. The real magic starts when we drop in something called a ‘seed prompt,’ and that’s about 15 minutes in. So if the early installs feel slow, stay with me — you’ll see why the install is worth it.”

“If the early installs feel slow, stay with me.” That sentence is the one that lets the anxiety drop. It tells the attendee: this part is supposed to feel like this. Don’t read it as the whole class. The good part is coming.

the meta lesson

In a hands-on AI workshop, the technical content is the easy part. The hard part is sequencing the anxiety.

That’s the meta-lesson from the dry-run, and it’s bigger than AI training. It applies to any technical session where you’re asking non-technical attendees to do hands-on keyboard work in front of their peers. The Excel power-user course. The new-CRM-rollout training. The new security-policy walkthrough. Every one of those sessions has an install or setup phase where the attendees are forced into “did I click the right thing” mode while the trainer is trying to be conceptual on stage.

You can’t remove the anxiety. But you can sequence it.

• Split the roles. Don’t ask one person to be both the conceptual presenter AND the per-attendee IT support. Two roles, two people. The presenter stays conceptual; the support person handles the keyboard work attendee by attendee.
• Promise the destination first. The opening slide isn’t “Agenda.” It’s “Here’s what you’ll walk out with in 60 minutes.” Give people a target to hold onto while they sit through the setup phase.
• Name the slow part. Tell people, out loud, “this next 10 minutes is the install phase — the real magic starts at slide N.” When the slow part is named, it stops being a problem. When it’s unnamed, it becomes the whole class.
• Find your inflection point. Every workshop has one — the moment where the abstract becomes concrete and the attendee can suddenly see what they’re getting. Identify yours. Label it. Make it the landmark you keep referring back to.

why this matters more in 2026

AI training is going to get rolled out at almost every company in the next 12-18 months. Most of those rollouts will be designed by people who are good at the technical content but haven’t thought hard about the workshop design.

The failure mode is going to be the one I just described: solid material, weak sequencing, the first 10 minutes lose the room, the attendees walk out remembering the install pain and not the agent.

The companies that get this right will treat the workshop itself as a designed product. They’ll have pit crews. They’ll have opening slides that promise destinations. They’ll have inflection-point landmarks. They’ll have practiced their dry-runs before the real rollouts.

And the people they train will actually use what they’re given — instead of forgetting it the moment the class ends because the dominant memory is anxiety.

This is a behind-the-scenes look at how I’m shipping an AI training rollout. The dry-run feedback above is real — the install-phase comment, the seed-prompt inflection point, the pit-crew suggestion all came from senior peers giving me honest read-backs. If you’ve rolled out AI training to non-technical staff and have your own dry-run lessons, I’d love to compare notes.