What We Never Get To

There is a section in every AI training session I deliver that I think of as The Good Bit. It is the part where we stop talking about policy and start actually doing something. It is the part I have prepared for, practised, and in several cases stayed up until midnight refining. It is, without exception, the part we never get to.

The Policy Ceremony, or: How Fifty Minutes Become Eleven

The session is booked for two hours. Two hours feels, in planning, like an embarrassment of riches, an extravagance of time, a full suite of possibility. In the week before, I have imagined what I will do with those two hours. I have built it in my mind like an architect: there will be a fifteen-minute conceptual foundation, a thirty-minute interactive exploration, a segment where participants try the thing themselves and discover, with something approaching delight, that it actually works. There will be a moment, I have planned for this specifically, where someone in the back row says “oh, hang on” in the particular tone of voice that means the penny has dropped.

What actually happens is that policy arrives first and sits down in the middle of the room and does not move.

Policy is not a person, technically, though by the end of a session it has developed something like a personality: cautious, repetitive, deeply concerned about data sovereignty, and constitutionally incapable of summary. The acceptable use policy takes twelve minutes. The data classification discussion takes twenty, because someone (and I say this with nothing but professional warmth) always asks whether their email signature constitutes sensitive information. The question of whether participants can use AI tools for student feedback loops into a parallel conversation about academic integrity, which begets a discussion of what constitutes plagiarism in the age of AI, which begets the question I have now heard in fourteen consecutive sessions: “So is it going to take our jobs?”

The jobs question is never a quick one. In theory, it should be. In practice, it expands to fill whatever time remains, because it is not really a question about AI at all; it is a question about mortality, relevance, and the specific anxiety of being an educator in a system that seems to be renegotiating its own contract with human expertise in real time. I have tried giving short answers. I have tried giving nuanced answers. I have tried the “let’s park that and come back to it” technique, which everyone knows is code for “we will not be coming back to it.” Nothing helps. The jobs question is the tidal force of every AI training session; it pulls everything toward it and only releases its grip when the room begins to pack up.

By the time we have navigated policy, addressed jobs, clarified that no, participants may not ask an AI to write their annual performance review in full and then sign it (this is specific to one memorable session and I will not be elaborating), and worked through the twelve-minute warm-up activity I designed to be “low stakes and generative” - we have eleven minutes left.

Eleven minutes. I look at my lesson plan, which has been running in the background like an optimistic background process, and I quietly close the tab.

What Lives at Minute 47, and Why It Matters That Nobody Sees It

Here is what the good bit looks like, for the record. Because someone should know it exists.

The good bit is where we stop explaining what AI is and start using it to do something specific and slightly ridiculous that nevertheless demonstrates a genuine principle. It is the moment where I give a room full of people a genuinely weird prompt and we watch the response together and I do not narrate it; I just let the room react. It is the part where someone asks “can it do that?” and I say “let’s find out” and we find out together, live, in real time, with the specific electricity of a shared experiment that might not work.

It is, frankly, what the whole session is for. The policy is scaffolding. The jobs conversation is important but hardly terminal. The warm-up activity is infrastructure. What I actually want people to leave with, the thing that changes how someone relates to these tools, that makes the difference between using AI occasionally out of obligation and using it with genuine curiosity - is the experience of discovering what it does when you take it somewhere unexpected. Of treating a prompt not as a search query but as the opening line of a conversation that has no predetermined destination.

I want them to understand that the prompt box is, more than anything else, a place where imagination has consequences. That the quality of what you ask shapes the quality of what you receive. That “boring in, boring out” is the first and most important principle of AI literacy, and that the corollary, “interesting in, interesting out”, is the one that will actually make them want to use it.

We get to none of this. We have eleven minutes, which is enough time to show three slides and ask if there are any questions, which there are, and which are, invariably, about policy.

In Which I Am Forced to Confess My Only Real Excel Achievement

This is the context you need to understand what happened last weekend: I have been asked to include Excel in my training delivery. Basic Excel. Practical Excel. Excel for people who use it for work, in the way that normal people use it for work: tracking budgets, organising rosters, making the kind of clean, functional spreadsheets that suggest competence without requiring any real inner life.

The difficulty is that I find Excel very boring. I am not proud of this. It is an objectively powerful tool with legitimate applications across virtually every professional domain. Many excellent people have Excel as a genuine skill. I respect this. I also cannot summon any feeling of engagement with it whatsoever, for a reason that is entirely my own fault.

The last time I was genuinely excited about a spreadsheet, I was building a Persian sentence generator.

I should explain. Farsi (Persian) has a beautifully deterministic word order. The grammar is rule-governed in a way that makes it, theoretically, amenable to systematic representation. The verb goes at the end. Always. The object precedes it. The structure is Subject–Object–Verb, with a tidiness that I found, at a particular point in an idle weekend, almost offensively elegant. So I built a spreadsheet. Not a small one. A spreadsheet with data validation drop-down lists for subject pronouns, a separate validated list drawing from a Farsi verb dictionary I had found and transcribed, conditional formatting that changed the cell colour depending on tense, and a sentence assembly formula in the final column that concatenated the selections into grammatically valid Persian output with the right-to-left rendering properly configured for the cell.

It worked beautifully. I used it approximately once, to show a colleague who expressed polite interest and then changed the subject. It has sat in a folder on my desktop ever since, a monument to what happens when you give a person with a linguistics background a copy of Excel and a quiet Sunday.

The point is: when I try to feel enthusiasm for showing someone how to do a SUM formula in column B, some part of my brain registers this as a significant professional downgrade from the Persian verb dictionary. Which is, I acknowledge, an entirely unreasonable position. Normal Excel training does not require a Farsi lexicon. The participants do not need to know that my personal relationship with spreadsheets peaked in a linguistic hobbyist fugue state at eleven on a Sunday night. This is my problem to manage.

The solution I arrived at, sensibly, professionally, using AI tools as a genuine training asset, was to ask Claude to help me build a Train-the-Trainer manual for Excel.

This is where it gets complicated.

Saturday Night and the Ozark Pivot: A Story About Prompt Engineering

I started, as one does, with a reasonable brief. I needed training materials. I needed them to be engaging. I needed them to make pivot tables feel consequential and VLOOKUP feel like a skill worth having. I am not, to be precise, bored by the concept of Excel. I am bored by the absence of stakes. Normal Excel training suffers from a dramatic problem: nobody cares whether the SUM formula is correct because the fictional budget in the exercise is fictional and everyone knows it.

What if the budget were not fictional? What if the stakes were, let us say, considerably higher than whether the quarterly variance figure looks tidy?

I cannot reconstruct exactly how the conversation arrived where it did. I can tell you that it began with me asking for a scenario-based training framework that would make Excel feel genuinely necessary. I can tell you that I mentioned wanting something with dramatic tension. I can tell you that at some point I wrote, in the prompt box, something to the effect of: what if the participant was someone who had to appear to be learning Excel for legitimate business purposes, while actually acquiring skills to manage financial records of a more complicated nature?

I would like to be clear that this is a pedagogically sound instinct. Scenario-based learning works precisely because invented stakes feel realer than no stakes at all. The research supports contextual embedding of skills. I was not, to be absolutely transparent, thinking about this research when I wrote the prompt. I was thinking about Marty Byrde.

What followed was approximately three hours of the most engaged spreadsheet content development I have undertaken in my professional life. We built a full manual. Session objectives that read like threat assessments. A case study involving a lakeside resort with inexplicably healthy cash reserves and a payroll that did not, on close inspection, correspond to any observable staffing. Practice exercises in conditional formatting specifically designed to highlight discrepancies before (as the manual delicately phrased it) “external stakeholders identify them first.” A VLOOKUP tutorial anchored to the scenario of reconciling two versions of the same invoice that had, through no fault of anyone, developed slightly different totals.

The Excel was technically impeccable. Every formula was correct. Every pedagogical scaffold was sound. The learning objectives were specific, measurable, achievable, relevant, and time-bound in ways that my standard Excel training had never quite managed, because my standard Excel training had never before had the structural advantage of organised crime as a motivating context.

I also commissioned a quiz. An Ozark-themed self-assessment with a dark aesthetic and amber typography and questions like: Your pivot table reveals a $47,000 variance in Q3 lakeside accommodation revenue. You have ninety minutes before the meeting. Which action is most appropriate? The answer choices were all Excel functions. The correct answer was legitimately the correct answer. The scenario was, to be clear, entirely fictional. I spent forty minutes on the visual design.

By Saturday night, I had a complete, polished training resource that I will almost certainly never be able to deploy in any professional context ever, because explaining the thematic framework to a room of MVIT staff would require me to say the words “money laundering scenario” in what is, technically, a workplace.

I am choosing to count this as professional development anyway.

A Diagnosis, of Sorts: What Imposter Syndrome Does When It Has a Prompt Box

I have been thinking about what this weekend actually demonstrated, and I have arrived at a conclusion that is either reassuring or deeply worrying, depending on your relationship with the concept of professional identity.

Imposter syndrome, in its standard form, is a paralysing thing. It is the voice that says you are not qualified, not experienced enough, not the right person for the room. It has cost more excellent educators their confidence than I care to enumerate. In its conventional operation, it produces avoidance: you do not volunteer for the panel, you do not submit the abstract, you do not raise your hand in the meeting where someone more credentialed is already talking about your subject with magnificent self-assurance.

What I have discovered is that imposter syndrome, when given access to a language model and left unsupervised on a weekend, does something entirely different. It goes feral.

The prompt box removes the social friction that normally keeps the imposter voice tamped down. There is no panel to be excluded from. There is no audience watching you claim expertise you are not sure you have. It is just you and the blank input field and the whole unguarded territory of your actual interests; which turn out, when the performance of professional credibility is not required, to be considerably stranger and more interesting than your lesson plans suggest. Nobody is watching when you type. Nobody will evaluate whether a Farsi verb dictionary was a reasonable use of Excel. Nobody will question the institutional alignment of an Ozark money-laundering scenario as a vehicle for teaching VLOOKUP.

You just do the interesting thing. And the interesting thing, it turns out, actually works.

This is, I think, what I am trying to give participants in those eleven final minutes, and failing, because the policy ceremony took the rest. Not the tools. Not the functions. Not even the prompt frameworks, though those matter. I want to give them the experience of typing something genuinely weird into a box and discovering that the box takes it seriously. That the thing they find boring can be made interesting by treating it as interesting. That expertise is not a fixed credential you either have or don’t; it is a practice of sustained curiosity, applied with enough specificity and enough nerve to follow the prompt wherever it goes.

I am, in this respect, a better practitioner on a Saturday night than I am in front of a room on a Tuesday afternoon. The audience in a training session makes me careful. The audience of zero, at a laptop, with a Farsi spreadsheet in the background and no particular outcome to justify, makes me interesting.

I am not sure what to do with this information. I am writing it down, which is probably a start.

The Part We Never Get To

On Monday I will deliver a session. The acceptable use policy will arrive promptly and take its seat at the front of the room. The jobs question will follow, as it always does, with the implacable timing of a recurring character who knows they are essential to the narrative even if they are not technically on the schedule. We will get to eleven minutes. I will look at my lesson plan with the specific expression of a person watching an excellent meal being carried past their table to someone else’s.

And I will think about the Ozark pivot tables, sitting in a folder on my desktop like classified material that has accidentally been formatted beautifully. I will think about the Persian verb dictionary and its perfectly conditional cells. I will think about what it would look like if we had a third hour, just one, just occasionally, where the policy had already been read, the jobs question had already been answered, and we could simply open the prompt box and see what happened when we stopped being professional about it.

I think it would be rather good. I think some of those people in the back row, the ones who’ve spent the session nodding politely at the acceptable use framework and waiting for something that feels real; I think one of them would type something genuinely peculiar into a box and discover, with something approaching amazement, that the box took it seriously.

I think that’s the moment that changes things.

We have eleven minutes. We’ll try again next session.