The thesis.

For thirty years, IT trained people not to touch. Automation reduced most people’s working computer to the browser and the inbox, and those of us in IT taught them to leave everything else alone.

AI platforms demand the opposite:

“You have to TOUCH ALL OF IT - and not fear the consequences.”

Enablement is not teaching a tool. It is un-training a learned helplessness that IT itself installed. Everything below serves that un-training.

And be clear about the scale. This is not a software rollout. It is not welcome to Word, or here is Excel. It is a business process transformation and a call for workers to think and perform differently - a preparation for a new world of work, in every way.

It applies upward as well. Don’t-touch runs deepest at the top, where delegation has made it invisible. Seniority is not fluency.

What gets taught.

The curriculum underneath the arc is not mine. I draw on the AI Fluency Framework by Rick Dakan and Joseph Feller. They developed the course with Anthropic. The classroom examples and the three moves below are mine. The AI Fluency Framework · The 4D framework course. In my room it comes out in plain words, as three moves:

Delegation. “Examine your workflow - where does AI slot in?” This is the moonshot question and the ladder’s third rung, asked before anyone has heard the word delegation.

Description. Learn to communicate clearly enough to deploy AI. The homework on paper - a story that gets your agent from point A to point B - is this discipline without the jargon.

Discernment and diligence. “Review your work and be ready to claim it.” I teach these as one move, because the review and the ownership are the same act. If you are not ready to stand by the output, you are not done reviewing it.

The ROI question.

What is the benefit of a good AI deployment - give me the figures.

Before the figures, a preface: most organizations don’t really have a choice. The technology is out of the box, and the reality now is enable or get left behind. But the investment is real, and it deserves honest accounting. Structurally, most organizations will need to upgrade technologically just to deploy this well. You may need outside help for strategy, implementation, and enablement. And people-hours will be spent learning, not doing.

So where does an organization see true, measurable financial gain?

Quality, capacity, better outcomes - and resilience.

Quality. Think of AI processes as another set of employees in the system. More eyes on a report before it goes out the door. More hands on a campaign before it goes live.

Capacity. An agentic workforce - developed by your staff, directed by your own workflows - multiplies the workforce you already have. More work is possible.

Better outcomes. The hidden benefit, because it is multifaceted. Employees who can work robustly and creatively in their domains - rather than on the administrative edges of them - work harder, are happier, and are more productive. The subtler side: AI doesn’t take the shortcuts we humans take, and it doesn’t carry our particular biases. It carries a different set of biases - not an absence of them - which is why the human stays in the loop. Lean on the deterministic side of agentic workflows and you get cleaner, more data-driven decisions, over larger sets of data.

Resilience. The quiet value-add. An agentic workforce is tireless at exactly the work no human team sustains: checking everything on the list it is given, not a sample - every document, every link, every permission on the list. Full coverage of a list does not guarantee catching every error; it guarantees the list was covered. Security has always run on impossible arithmetic - the attacker needs one hole, the defense needs to find them all - and exhaustive, boring search is what agents are built for. In November 2026, AI-assisted research found a major security flaw in Java, a software platform much of the internet stands on. The same pattern works at the scale of one organization: agents that reread the report nobody rereads, revisit the permissions nobody revisits, retest the workflow nobody remembers building - and surface the frailty before it becomes the incident.

And the clock runs both ways. The attackers have agents too. Legacy systems that survived on obscurity - too boring to probe, too tangled to map - are now porous to tireless search. Updating them stops being a someday project. That is part of the honest accounting above.

These gains are measured as capacity gained and work made better - not headcount to cut. The pricing firewall below keeps me from pricing my own claims. It does not decide what management does with the result. So before the baseline is taken, the sponsor answers one question out loud: can the hours people report be used to justify cutting a job on their team? If the answer is no, say who sees the data and how that holds. If leadership will not rule it out, say that too. People who are denied an answer protect themselves by keeping quiet - and then nobody measures anything.

The arc.

Ten steps, three disciplines, one list.

Prepare · 01-05 Begin · 06-07 Continue · 08-10 Running the whole time · A-C

Before the classroom

01 The survey.

On day one, send a short survey across the whole organization: ten questions or fewer. Have you used AI? Do you want to? Which tools? Where did it work, and where did it not?

The tenth question:

“If this AI stuff turns out to be the crazy cool thing that actually does make work better, what is that moonshot project?”

The first nine take the temperature. The tenth recruits. It plants a seed that returns months later as the homework in step seven.

The baseline rule. Before anything is rolled out, everyone writes down how long their tasks take without AI. Unless this is agreed before implementation, it will not be done evenly or thoroughly, and the “before” picture cannot be rebuilt afterward. Measurement begins here.

02 The sort.

Three piles, two levels.

Sort the individuals into three piles: for it, against it, and already using it.

Then meet every department head face to face, ask the same questions, and sort the departments into the same three piles. Now there are two maps.

Where the maps disagree, record the mismatch and return to it once the first work is under way. Know them. Log them. Do not lead with them.

03 Governance first.

Where a department’s resistance is about the rules, strengthen the rules. Treat the objection as a list of things to get right. Those departments more than likely have valid concerns. Come prepared to talk and to take notes.

There are four things the organization has to know about itself. None of them is about technology.

  1. Does our data go straight into AI platforms, or through a layer in between?
  2. Is AI-written content for outside audiences acceptable, or fundamentally opposed?
  3. Is the first goal to take administrative work off people, or something more?
  4. Is there an acceptable-use policy that everyone understands: what goes in, what comes out, and who is responsible?

Answer the objection the person actually raised. If the answer is still no, I don’t need a different reason.

A person can understand the rules and still judge a use not worth doing. That no gets recorded and honored - the same as the one in discipline C.

04 The timeline.

The organization states what it wants from AI now, in a quarter, in a year, and in five years, with a pathway to each.

Go to the holdouts only when the rules and the timeline are both in hand.

05 The department heads.

This comes before any classroom. The department heads must be won over. If they are not, their teams will not succeed. A team cannot out-adopt its own boss.

The conversation has three beats, in this order.

Empathy. Ask why the resistance, keep the conversation relaxed, and listen. Expect news stories, laid-off nephews, Twitter. This is not a conversation for preaching.

Inspiration. Then evangelize. The technology is remarkable. Take what is frightening and make it exciting.

Inevitability. “this is the direction of the company, so this is indeed now your direction as well.”

The third beat is the soft hammer, and it lands only after the first two. Delivered cold, it is a threat. After real listening, it is gravity.

The messenger.

I taught my first Welcome to Word class at twenty-six, in a new suit from a discount retailer, in front of a full class of experienced secretaries. None of them were interested in learning from me. I didn’t know the particulars of their world well enough to know which parts of the technology were worth highlighting. And my very presence was upsetting: a young upstart telling them how to do a job they had been doing for my entire lifetime.

All of those things were real. Maybe not valid, maybe not ideal - but real, and they are real considerations for a successful deployment. I can still see that classroom.

So: whoever is part of the deployment needs to be relatable, and needs to have empathy. The listening in the department-head conversation is not a tactic. It is what I owed that room of secretaries and didn’t yet know how to give.

In the classroom

06 Wheels up.

Session one.

Everyone gets an approved place to make a first artifact. In my engagement, we used mail and calendar. In your room, start with a practice inbox unless the data owner has approved that use of real mail. Use the memory and retention settings your organization has approved; if that decision is open, leave memory off. No message is sent and no calendar entry is changed without a person approving it.

The uniform setup is not optional. Identical doorframes are what allow every door to be different.

If that sounds at odds with opting in, it is not. Everyone gets the same tools, and everyone uses them in their own way. The installation is uniform. The use is voluntary.

The sequence:

“Open. Make a project. Put a chat in the project. Test the connectors with a quick build. Make that first artifact. Be amazed.”

A connector is the approved way the tool reads information from another application.

The standard first build is My EA: mail and calendar.

Misfires are part of the lesson. Connectors will misfire, and each misfire is a chance to teach troubleshooting during the session. It broke, you touched it, you fixed it, and nothing bad happened.

07 The moonshot homework.

“I just gave you the keys to the castle - what is that BIG project you couldn't do last year?”

On paper. Start with the big picture: a beginning, a middle, and an end. Fill it in as you would a short story. Then test it: would someone who does not know the story arrive at the same place? Keep revising until they would.

Where it leads:

“All you are doing is giving a story for your agent to get from point A to point B.”

That is writing instructions for a machine without ever calling it that. It is the Masterman half of the name: ordinary language, written well, drives the machine.

After the classroom

08 The ladder.

A four-point scale. Every user stands on one of the four points.

  1. Onboarded. The seat is secured, the software is installed, the connectors are up, and the settings are correct.
  2. First skill. Usually completed in class. Here the user begins to understand the import of natural-language coding.
  3. First independent project. Assignment-driven, with weekly check-ins and support. The assignment builds on the moonshot: “now that you know a bit about how these things work, what is the part of your workflow that isn't automated that you think you can automate now.” Not the castle. The first corner they can automate, chosen by them.
  4. Beyond chat, onto code. After the first project, some strong users need the next layer of tools. I assess this by the scope of their next project and by the questions they ask: How do I do this faster? Is there a good way to do this hands-off? Readiness is listened for, not tested.

Above the scale there is a longer arc: adoption, then ownership, then stewardship, where the community looks after its own skills and I hand over the keys.

The entry door.

Nobody adopts AI in general. Each person comes in through one door: the task they already hate. Two people from the engagement behind this site.

One owned reporting that ate a day at a time. His door opened on two realizations at once: this turns the day’s report into a click - and, you have to check it, you have to keep it working, and sometimes you have to talk back to the robots. He became a star user because he held the vast capability and the limitations in the same hand, and he wasn’t put off by working with the tool as a colleague alongside him. At one point the tool told him, “I don't have that tool.” He knew it did. “Actually, you do - go check.” It did. A small example, but it shows the thing that matters: the user’s willingness to exercise diligence, to iterate and refine until the result is acceptable.

The other ran a call center and hated the headphones - hours of grading calls by hand. She was the one who said, “I know AI can help me with this.” Once she adopted the tools, the change was physical. Her confidence returned. She was no longer overwhelmed by the drudgery of her job, and she started dreaming bigger in her role: combining her notes into weekly newsletters for her remote staff, partnering on the reporting, recoding the ops-center call-QA tool. Because she could immediately identify where AI would take away the parts of the job she didn’t like, she was immediately transformed. It brought her hope.

The method’s job is to find each person’s door. The survey’s tenth question, the moonshot homework, and the ladder’s third rung are the same question in different clothes: what do you hate doing that this could take?

09 The champion.

Succession begins in week two. There is usually a student who is ahead of the class and wants to share that advantage.

“Thank you. You made that so much fun, and I learned something from you today.

Could I ask for your help in your department?

Nothing formal - would you be willing to be someone people can come to with their questions?

And I have been thinking about a Teams or Slack channel for AI user questions. If we set one up, would you help me in it?”

No title, no anointing, nothing to resent.

This happens early, while people are still climbing the ladder. It is listed ninth because it is what makes the leaving possible.

The end

10 The leaving.

The engagement ends. The relationship changes state. The parting speech:

“You are now my AI colleagues - I know you now - we're in this field together. Send me questions. I'll send you questions. This is a learning community that transcends org boundaries. That is the cool thing about a new tech frontier - we are all dauntless explorers right now. In 5 years, we will look back and be glad that we collaborated after this engagement.”

I am comfortable leaving when three things are true.

  1. Management is excited and keeping it alive.
  2. There is an active Slack or Teams channel where people give and get help without me.
  3. The cross-department introductions are already made. People are connected to each other, not to me.

The horizon the whole method points at is the data step. AI inside one department is useful. AI across separate departments, unifying their workflows, is the goal. That is why I make sure the cross-team partnerships exist before I go.

The six-month bar.

This is a new frontier. Not many organizations are here yet, so some of this is prediction and some of it is hope. Six months after wheels up, here is what I hope to see:

  1. Everyone has hands on the tool, and their administrative responsibilities are automated.
  2. Style and branding guidelines are set up and automating the external-facing documentation.
  3. Every external-facing URL has been through an AEO audit - a check of how AI systems read your public pages - and the organization is ready for the agentic future of web crawling.
  4. And with all this boring stuff out of the way: the organization is working cross-departmentally on shared datasets, looking for insights that hadn’t been surfaceable until now.

The fourth is the point. The first three buy the time for it.

Running the whole time.

Three disciplines do not wait their turn. They start early and they never stop.

A The replaceable talk.

Never tell people AI will not replace them. They know a soothing noise when they hear one. This is what I say instead:

“Replaceable is always on the horizon - always. We're all replaceable. But you are a professional, you are still working, you know this. What makes someone retain their job is their willingness to adapt, to grow, to learn, to add value to themselves and to the org. Still, you are replaceable. The goal here is to stack the odds in your favor.”

For the right listener there is a second beat: “Adopt the efficiencies. Use this time as a headstart over others.” Not everyone gets that one. You have to pick the right people for it.

There is a second fear under the first: that they will break something and be blamed. IT policy taught people that one.

B Measurement and the ledger.

Measurement is key. It starts at step one with the baseline rule and it never stops.

Every interaction is data: emails, Zoom calls, direct messages, handwritten notes uploaded. People used to compile all of that by hand. Now they do not. A second set of eyes, human and AI. The way I measure is the method itself, demonstrated.

Records someone agreed to contribute, not conversations that merely exist.

If the sponsor cannot rule out using identifiable time estimates for staffing decisions, I do not collect named hours for this exercise. A person may keep their own estimate private or skip the exercise.

What happened in the field. In the engagement behind the figures on this site, the sponsor asked at week ten what the work was worth. I went back through every weekly update I had sent and priced each deliverable myself, counting only what I had already reported. That produced $194.5K-$387.5K in delivered value by week ten. It was reconstructed, by me, looking back. My note-taking saved me. That is not good enough.

Two rules came out of it.

  1. Keep the ledger from week one, with each department, so the value question is answered as the work proceeds.
  2. The pricing firewall. What I tell the sponsor: “I am not reporting dollar amounts - I will give you time saved and the person where it was saved - YOU compute the dollar amount.” I certify the facts: the hours, and whose they were. The organization applies its own rates to its own people. I never hold the multiplier, so I cannot inflate the number.

The dollar range above is my own look-back estimate. The firewall is the method going forward. They are not the same thing.

The human cadence is simple: check in, nudge, listen for the big wins, and help with the roadblocks.

The culture it installs: build in public, build through iteration, ask questions, be vulnerable.

C The intervention and the no.

The tripwire is two weeks of silence. Then a Teams message:

“Hey, how's your clauding.”

If the answer is good, I go one gentle step further: “what are you working on, need help, would love to see - I always learn from others' builds.” The learner’s chair, never the auditor’s.

If the silence continues: a desk visit, a phone call, a Zoom. Each step moves toward the more human channel.

The calibrated gift. Used sparingly, and only with people I know will receive it well, I build something for the stalled person and post it publicly: “hey, I'm saying hi with my Claude build for you.”

The no. One person received the gift and a visit in person, and still did not want to do it. It did not sing to him, and I left him be.

Opt-in that cannot be declined is a mandate with better manners. The opt-in rate is credible because it is not 100 percent.

The responsiveness rule. This work demands a fast reply, even when that means answering at 10 PM. For the people who are scared, the 10 PM reply is the intervention.

What never works.

  1. Teaching someone to code over Zoom. Zoom calls do not get people where they need to be.
  2. Handing out documents and videos and expecting people to read and watch them.
  3. Setting up connectors over Zoom, unless the person is unusually engaged.
  4. Telling people they must use the tool or be fired.
  5. Slow response. Be very responsive, even at 10 PM.
  6. Technical jargon. The thing being taught is natural-language programming, so use natural language. That is the whole point.
  7. Prompt libraries. Make them if you like, but I tell people not to use them. They hamper growth in this technology.
  8. Being a jerk. People are scared.
  9. Inventing an answer on the spot. Say “I don't know.” It is fair to add that the field is moving fast and you will have to check. People are very sensitive to invention.

A published method is itself a kind of recipe, so number seven applies to this page. My answer is that a practice can be standardized without an approved script. The steps are the practice. The words in the room are never scripted.