Train & enable
Most AI training is a demo followed by a deck. Ours is a working session: your people build assistants for their own jobs, on the platform you have already standardized on, and walk out using them. Run by the engineers who build production AI systems, not by presenters who read about them.
The order is the method. Most training starts at module four, which is why the enthusiasm dies in week two. Most of the clock is people building; we talk when it unblocks someone.
Module 01
The desktop app rather than a browser tab. A project per initiative. One shared place the tool can read your standing context from. Unglamorous, ten minutes, and the single biggest difference between the people who get value out of AI and the people who conclude it is overhyped.
Module 02
One saved block carrying your role, your organization, what you are accountable for this quarter, and what a good answer from it actually looks like. Then a smoke test that proves it took, before you trust it with anything that counts.
People who skip this spend the rest of the year re-explaining themselves every morning and wondering why the output reads like it was written for someone else.
Module 03
Take one job you would actually hand over. Build the check first: three real examples from your own work, what a good answer had to contain, and where the tool falls short today. Then score it honestly and find out where it is unreliable while that is still cheap to find out.
This is how we build production AI, scaled down to an afternoon. It is also the module nobody else runs.
Module 04
Turn the checked job into something reusable so you stop re-prompting it from scratch. Then aim it at real work: commitments pulled out of a meeting, a premortem on a decision that is still reversible, a Friday review that is about your actual week.
Plus the rule for what you paste into these tools, which is what keeps the other three modules from becoming an incident.
No attendance records to brag about yet. These are the constraints we run to.
Same method, different amount of room to work in. We will tell you on the call which one your situation actually needs, including when it is the cheaper one.
| Executive sessionHalf a day | Team buildEight hours · two mornings | |
|---|---|---|
| Room | A leadership team, however many that is. | A room of up to twenty-five, everyone building. |
| Shape | Where AI changes your economics, what has to be governed before it spreads, and one assistant built in the room. | Two mornings, laptops open: understand and operate first, then multiply and automate. Every module ends with a checkpoint on each person's own machine. |
| You leave with | Standing context that works, and a shared picture of what you are approving. | A configured, verified environment, two working skills you built, a scoped automation, and a company-level AI opportunity nominated for a pilot. |
| Best when | Your leadership team keeps approving AI spend without a shared picture of it. | A function wants the tool in daily use rather than on a wishlist. |
Scroll the table sideways to compare →
We teach on Claude or ChatGPT, whichever your organization has approved. We are deepest on Claude, and we would rather say that than claim equal expertise in every tool on the market. Standardized on Copilot or Gemini? Raise it on the call — most of the material carries over, and we will be straight about what does not.
Everyone walks out with a personal prompt library written around their company, their role, and what they are trying to get done this quarter. Not a tip sheet. The prompts they built and ran in the room, on their own work, ready to paste.
It is built module by module as the day climbs — and it is the follow-up habit: three prompts from it on real work in the first 72 hours is where adoption either sticks or does not.
// the highest-value follow-up on the sheet
Before I use that: where is it most likely wrong?
Give me the three weakest points in what you just
told me, ordered by how much damage each would do
if it turned out to be false. For each one, name the
specific thing I could check in under ten minutes
that would settle it.
Then tell me what you left out because I did not ask.
Most AI training is delivered by people who have never shipped an AI system. It shows: the demos work, the advice is generic, and nothing survives contact with a real workflow two weeks later.
We build production AI, so the session is drawn from systems we have actually shipped, including the parts that failed.
Anyone can demo a clever prompt. We spend a whole module on how to tell when it is quietly wrong, because that is what makes it safe to rely on.
We work on the platform you have approved. Training does not need your confidential data, and we retain nothing afterwards.
A baseline before, a follow-up after, and a check on whether what people built is still running a month later. Adoption is a habit problem, not an information problem.
If the gap is a missing system rather than a missing skill, you want engineering, and we would rather scope that honestly than sell you a workshop that cannot fix it.
Every module ends with a checkpoint on each participant's own machine. You watch it working before the room moves on, not on the trainer's laptop.
Training gets a team fluent. It does not build you a system. When the next step is engineering rather than enablement, it is the same people either way.
Two things people usually want to know that are not really FAQ material.
What it costs. One number for the session, no per-seat math — quoted on a short discovery call once we know the format and who is in the room. The call is free, and you get a candid read on whether training is even the right spend. Contact us for pricing.
What happens if it does not land. We baseline the group beforehand and survey afterwards, but the measure that matters is whether the assistants people built are still in use a month later. We ask for that too, and we would rather find out it did not stick than not ask.
Up to twenty-five in a build session — enough to put a whole function through it at once rather than a pilot group. We bring the structure and enough hands to keep a big room building instead of watching. More than twenty-five to train? We run multiple sessions and sequence them.
Both. Remote sessions are run live, not pre-recorded, and distributed teams do fine. In person is better when the goal is to shift how a leadership team thinks rather than to teach a workflow.
Not always. Some of the session works on free tiers, and some of it, particularly projects and reusable assistants, needs a paid seat. We tell you exactly what the agenda requires before the session so procurement is not a surprise on the day.
Training does not require your confidential data. We work from public and internal-but-non-sensitive material, take least-privilege access when a system has to be touched at all, and retain nothing after the engagement. The session includes a rule for deciding what is safe to paste into an AI tool.
Yes. Every session includes a practical filter for what goes into these tools, and we will review an existing AI policy or help draft a first one. Deeper governance work, evaluation harnesses, guardrails, and audit trails, is our consulting practice rather than a training module.
Their own standing context, a checked assistant for a job they repeat, and a written check that tells them when that assistant is getting it wrong. Plus a personal prompt library from the day, adapted to their company and role.
Thirty minutes to work out the format, the platform, and whether training is even the right spend. No pitch.