Sebbe Jones

The team workshop

Your Judgment, at Machine Speed.

Your team already uses AI. What nobody trained is the other half of the job: choosing the right work for it, giving it what it needs, improving what it returns, and knowing how hard to check before the work goes out. Three hours on their own tasks, or ninety minutes when that is what the calendar allows, and everyone leaves with the same method.

Let's talk about your team Or write me and we can find a time: sebbe@sebbejones.com

The problem this solves

Two teams, same tool, different results.

AI arrived in most organizations without training and without rules. People worked out privately what it was good for, which means the skill level inside one team can vary more than the tools do. One person is saving an afternoon a week. The person at the next desk pasted in a two-line prompt, got something generic, and quietly went back to doing it by hand.

Both of those are expensive, in opposite directions. The first is fragile, because the method lives in one person's head. The second is time you are paying for and not getting back.

Then there is the part everybody worries about, and they are right to. A wrong answer arrives in the same polished, confident tone as a right one. If nobody has agreed on how much checking a piece of work deserves, the decision falls to whoever happens to be in a hurry.

This workshop replaces private habits with a shared method, on the team's own work, in one sitting.

Who it is for

Teams whose output reaches somebody who matters.

Operations, client service, finance, marketing, anyone whose work ends up in front of a customer or a decision-maker. The common thread is not the department. It is that judgment is part of the job, and the work carries the organization's name when it goes out.

No technical background is required and the method assumes none. It is built for one team, small enough that everyone works on their own task rather than watching somebody else's.

The full session

What the team actually does.

The session runs in order. Each part uses the one before it, and the team's own tasks carry through from the first exercise to the last. This is the full three hours. The shorter formats are cut from it, and they are described further down.

  1. Build a working mental model

    We start with what the tool is doing when it answers, in plain language, no technical detail for its own sake. The point is prediction. Once people can guess where it will be strong and where it will be thin, they stop treating every task the same way. This is what closes the gap between the person who refuses to use it and the person who trusts it flat.

  2. Sort the work before touching it

    Everyone arrives with recurring tasks they actually do. We put them on the Judgment Grid: how much time AI saves here, and what it costs if the answer is wrong. Some tasks come off the list entirely, which is a result. You are not sorting tools. You are sorting decisions.

  3. Give the tool what it does not have

    Most disappointing output is a context problem, not a capability problem. The tool does not know your policy, your client, your standards, or what last quarter looked like. The team takes one real task and rebuilds the request around what a competent new hire would have needed to be told.

    Exhibit: the same request, with context
    Before: "Write a reply to this customer complaint."
    After: "Reply to this complaint. Our return window is 14 days, no exceptions under 30 days without a manager. This customer has ordered four times. Match the tone of the two replies below. Do not offer anything I have not listed."

  4. Ask well, then treat the answer as a draft

    DRAFT is the core of the session, and the team learns it by running it on their own work rather than watching me run it on mine.

    Read what comes back as a draft. Push on it. Say what is wrong with it and ask again. Edit it yourself where editing is faster than asking. Then the sign-your-name test, which is the only quality bar that has ever worked: would you put your name on this as it stands?

    D
    Define the outcome
    R
    Role
    A
    Add context
    F
    Format
    T
    Test and tighten

  5. Match the check to the stakes

    Checking every line gives back the time you just saved. Checking nothing puts a wrong number in front of a client. Between those is a decision the team can make deliberately, based on where the task landed on the grid. We work on where to look first, and the answer is usually the seams: the places where the tool was filling a gap in what it had been given. Names, numbers, dates, policies, anything that sounds specific and was never supplied. When something matters, it gets confirmed somewhere else, against the real source, a tool built for that job, or the person who would know.

    The output is telling you a story. It is built to be coherent, not correct, and nobody is hiding anything, which is what makes it hard: there is no tell, and the invented line arrives with the same confidence as the true one.

  6. Build it, then break it

    Small groups build an AI-assisted version of one of their own recurring tasks, end to end, then attack it until it produces something they would not send.

    The capstone

    Every group finds a failure. That is the exercise working. Finding it in the room costs an hour. Finding it later costs a client relationship.

  7. Keep what worked

    The session ends by making the day durable. A prompt that worked once gets written down. A written prompt used the same way twice is a process. A process the team trusts is what a system can eventually be built on. Document, then process, then systems, in that order, because teams that reach for the system first end up automating something nobody had agreed on yet.

One clarification

The high-stakes work is the center, not the whole session.

On the grid, the interesting square is the one where AI saves real time and a mistake is expensive. That is where customer replies, reports, and numbers that inform decisions sit, and it is where the Build-and-Break capstone happens.

But the session is wider than that square. Sorting the work comes before it, and the team has to decide what belongs there at all. Documenting and systemizing comes after it, which is the part that stops the value walking out with one person. A workshop that only taught careful checking would leave the two ends undone.

Low cost if wrong
High cost if wrong
Saves real time
Low cost if wrong Use itDrafts, first passes, reformatting.
High cost if wrong The centerSpeed pays, and a person stays in the loop. Build-and-Break happens here.
Saves little time
Low cost if wrong Skip itIf the prompt takes longer than the task.
High cost if wrong Keep it humanExposure, without much bought back.
The grid, compressed. The green square is where the capstone happens.

A fair objection

"Write the process down once and let AI run it."

A lot of AI training lands here, and the first half is right. Writing the process down is exactly what the last part of this session does.

Trusting the run without a check is where teams get burned. The output still has to be measured against your standards, and how hard to measure it depends on what the work is. That check is a skill. It can be taught, practiced, and made consistent across a team, which is most of what this session is for.

What is included

More than the hours in the room.

The session is the middle third. The work before it is what makes it yours, and the work after it is what decides whether any of it is still running a month later.

Before

Built around your work

  • A conversation about the team, the work, and what you want out of the day.
  • A short set of questions the team answers anonymously.
  • The examples and exercises rebuilt around what comes back, plus your AI policy if you have one.
In the room

Three hours, or ninety minutes

  • Hands-on throughout, on the team's own recurring tasks.
  • Four one-page handouts, given out as each part of the session arrives.
  • A starter pack of prompts, written the way the session teaches, so nobody is staring at a blank box on the Monday.
After

So it does not fade

  • A short feedback survey, with the results shared back to you rather than kept.
  • A recap within 48 hours, written so it can be forwarded to whoever approved the session.
  • A check-in from me at 30 days, to find out what stuck and what did not.
Nothing to buy and nothing to install. We work inside the AI tools your organization has already approved.

What the team leaves with

Ten things, all of them made in the room or sent after it.

This is the three-hour list. The 90-minute session produces nine of the ten: there is no room to document the process before they leave, so that one goes home as homework.

Made in the room

  • The team's own tasks placed on the Judgment Grid, with the oversight decisions already made.
  • A prompt for a real recurring task, with the business context built in.
  • An output they pushed back on, edited, and checked.
  • The start of a shared prompt library, so a prompt that works stops being one person's private advantage.
  • One AI-assisted process, built and stress-tested in the capstone.
  • One process worth documenting, written down while it is still fresh.

Materials and follow-up from me

  • Four one-page handouts, built to sit next to a keyboard rather than in a folder.
  • A starter pack of prompts, written the way the session teaches, for the tasks the team did not get to in the room.
  • A recap within 48 hours, written so it can be forwarded to whoever approved the session.
  • A check-in from me at 30 days, to find out what stuck and what did not.

Customization and intake

The session is built after we talk, not before.

Ahead of the day I ask you and a few people on the team what the work is, what they already use AI for, which recurring tasks are worth the time, and what your policy allows. If there is a written AI policy, I want it. If there is not, that comes up in the session, because the team will ask.

The structure of the session holds, whichever length you pick. The examples, the exercises, and everything the team writes down are yours. That is the difference between this and a session spent watching generic examples. There are demonstrations and I show my own work on screen, but the material the team works on is their own.

Formats

Three ways in. Same method.

The shorter versions are cut from the full session rather than written separately, so nothing contradicts anything.

  • Three hours is the full session, everything described above. It is the one I would pick if the calendar allows it, because the last part, where the team turns what worked into something repeatable, is the part that outlives the day.
  • Ninety minutes runs the same order and keeps the capstone, with one fewer exercise and no break. The one thing it does not produce is the documented process. Choose it when time is the constraint, not budget. It costs a little less, not half.
  • Fifteen minutes is for leadership. No laptops. What your people are almost certainly doing right now, what a confidently wrong answer costs, and how to decide which work to trust AI with. It ends on a decision about whether to bring it to the team. No charge for it.

Logistics

The practical part.

  • Three hours as one block or split across a morning, or the 90-minute session in one sitting.
  • In person, or over video with breakout rooms standing in for tables.
  • We work inside the AI tools your organization has already approved. Nothing to buy, nothing to install, no admin access needed.
  • Whatever your AI policy says, we work inside it. Including the parts that say do not put that in there.
  • Everyone needs a laptop and access to the tool they normally use.
  • Everyone brings one recurring task. That is the only preparation.
  • Pricing depends on the size of the team and whether the session is in person or remote, so I quote after we have talked.

Why me

The job was deciding what to trust. Then teaching it.

My job was pattern recognition with money on the line: spotting the thing that looked completely legitimate and wasn't, fast enough to act before it cost something. That started in 2002, in fraud prevention, when the technology had arrived and the rules had not been written yet.

The part that matters here is what came after. I built the teams that made those calls, trained the people, and wrote the processes so the judgment didn't live in one person's head. That is what this workshop does, in a domain where the confident wrong answer arrives as a paragraph instead of a transaction.

The full background →

Start here

Start with fifteen minutes.

The way in is a short conversation, not a booking. Fifteen minutes with whoever decides, no charge and nothing to prepare, ending on whether this is worth bringing to the team. If you already know you want the workshop, say so and we will talk about that instead.

Your name, your email, and one line about the team is all the form asks. I read them myself and reply within one business day with a few times that could work.

Let's talk about your team Or read the FAQ first.