The setup
Seventy of us. One prompt. The same week.
A controlled test, almost too tidy: a room of working professionals marched through the same five questions about AI, in the same week.
I went in certain that the thing that mattered about my classmates’ AI talks was the industry each of us spoke for. That was the mistake — and the embarrassing part is I’d already written it, in my own words, on one of my own slides.
Ask anyone how they know whether they’re keeping up at work and they’ll point sideways, at their own industry. The dentist watches other dentists. I’ve spent years in IT and quality assurance measuring myself against other IT people — same reflex, same lane. Then I read everyone else’s slides and the reflex fell apart in my hands.
The label that wouldn’t stay on
The setup was almost too tidy: a room of working professionals, a couple dozen industries, all marched through the same five questions about AI in the same week — same prompt, same rubric, same deadline. So I did what anyone would: I took them one industry at a time. Dentistry, pharma, utilities, national defense, insurance, recruiting, hospitality, IT. It looked like a map of the economy.
And it was the wrong map. The labels wouldn’t stay on. The moment I stopped reading the industry on the title slide and started reading what people actually said the tool did, everything rearranged itself in front of me — and it did not rearrange by industry.
The breadth
About two dozen fields. As far apart as fields get.
The room spanned regulated frontline work to creative and mission-driven roles — and it still funneled to the same handful of tools.
The first move
I read them one industry at a time. The labels wouldn’t stay on.
The moment I stopped reading the industry on the title slide and started reading what the tool actually did, everything rearranged itself in front of me — and not by industry.
By industry · the title slides
dentistry pharma utilities defense insurance recruiting IT
Looked like a map of the economy.
By what the tool actually did
screening drafting summarizing forecasting
It rearranged by tool — then by verb.
Everyone had packed the same tool
Here is the thing I did not expect. Nobody had built “AI for dentistry” or “AI for insurance.” They had all reached for the same three or four general-purpose tools and pointed them at their own work.
The finance analyst built her slide around ChatGPT and Microsoft Copilot. The manufacturing lead described “SOP copilots” on the shop floor. The organizational-development consultant had a “Comms Copilot” that drafts and a human that calibrates tone. The entertainment strategist had a “Writers’ Room Copilot.” The nonprofit operator ran what she called a “co-pilot / co-analyst.” Five fields that never met, each dressing the same tool in a local uniform and reaching for the same word for it: copilot.
That was the tell. Nobody had coined the word copilot — it’s a product name, Microsoft’s and GitHub’s, that had soaked so deep into how we talk that a dentist and a screenwriter both reached for it without thinking. We say “copilot” for the same reason we say “search” or “google it”: the tool named itself, and we went along. Which is the quiet proof of the costume — the dentist wasn’t competing with some specialized dental intelligence. She was pointing the same general model at an X-ray, and calling it by the same vendor’s word I did.
The tell
Five fields that never met. One word for the tool.
Nobody built “AI for dentistry.” They each dressed the same general engine in a local uniform — and reached for the same vendor’s word.
Under the costumes, the same few verbs
Once the tools collapsed, the tasks collapsed too — into a handful of jobs wearing two dozen sector nouns.
One was screening: look at a pile too big for a person and surface the few items a human should actually touch. The dentist flags decay on an X-ray. A chemical-QA lead flags the test result that doesn’t fit — she figured the tool could hand back two-thirds of her team’s investigative time. A recruiter ranks a résumé stack. An insurer scores a claim for fraud. A data-center tech ranks incidents by severity. Five industries, one verb, and it happens to be my verb too: quality work is triage.
But screening is the easy case, so here is a harder one that behaves identically: drafting the first pass a human then fixes. An HR lead generates the job posting. A consultant generates the first-cut analysis. A communications manager drafts the announcement. The screenwriters’ room spins up alternate scenes. Marketing produces the campaign copy. Change the noun on the front of the sentence and the slides are interchangeable — same tool, same verb, same handoff, five more costumes.
And the tool and the task list are one fact seen twice. The verbs collapse across industries because every industry handed them to the same general engine — and that engine happens to be fluent in exactly this short repertoire: screen a pile, draft a first pass, summarize, forecast, schedule. The industry was never the unit; the tool, and the handful of verbs it’s good at, is.
Under the costumes
Two dozen industries. A handful of verbs.
Once the tools collapsed, the tasks collapsed too. Change the noun on the front of the sentence and the slides are interchangeable.
Verb 1 · screening
Surface the few items a human should actually touch.
Dentist — flags decay on an X-ray
Chemical-QA lead — flags the odd result
Recruiter — ranks a résumé stack
Insurer — scores a claim for fraud
Data-center tech — ranks incidents by severity
Verb 2 · drafting
Generate the first pass a human then fixes.
HR lead — generates the job posting
Consultant — first-cut analysis
Comms manager — drafts the announcement
Writers’ room — spins up alternate scenes
Marketing — produces campaign copy
The engine’s whole repertoire
screen · draft · summarize · forecast · schedule
The slide where I fooled myself
I wasn’t an observer of this mistake; I was its clearest case.
I presented as “the IT industry.” My headline tool was Microsoft Copilot — I made the case that it was most companies’ on-ramp into Microsoft’s world, the way Amazon Q is the on-ramp into AWS. A few slots away in the same roster, the finance analyst had built her deck around the same Microsoft Copilot. We never spoke, and we drew different morals from it — I dressed the tool up as organizational citizenship, she listed it beside Bloomberg and FactSet. But we’d built two “industry” talks on one identical piece of software, each convinced we were briefing the room on our own field.
And the evidence was sitting on my own first slide the entire time. I’d written that the convergence was happening “primarily in the IT industry, irrespective of domains.” Irrespective of domains — the claim and its own refutation in one breath, and I filed the slide under IT anyway.
The slide where I fooled myself
Two “industry” talks, built on one identical piece of software.
The shared headline tool Microsoft CopilotMe · “the IT industry”
Dressed it up as organizational citizenship — the on-ramp into Microsoft’s world, the way Amazon Q is the on-ramp into AWS.
Finance analyst
Filed it beside Bloomberg and FactSet — a piece of the finance tooling stack.
On my own first slide: Convergence “primarily in the IT industry, irrespective of domains.”
So who am I actually up against
This is where the reflex breaks for good. A shared tool would mean nothing if it were Excel — everyone has a spreadsheet open, and no accountant is on anyone else’s curve because of it. But this isn’t Excel. Excel doesn’t improve at my job when someone in another field uses it; this one does — it’s a single shared model that gets better from everyone’s use.
The model the finance analyst and I share is one engine that improves release over release, pulled forward by how everyone leans on it. When underwriters and radiologists and recruiters push it harder at screening, the next version is better at screening mine — unevenly, on the labs’ schedule not mine, but in a direction my industry doesn’t set. So my future isn’t decided by the other IT people I’ve benchmarked against for years; it’s decided by what the shared engine gets good at next, driven by everyone leaning on it — most of them nowhere near my field. On the axis that ends up mattering, the finance analyst and I are on the same curve. The IT colleague who barely touches the tool isn’t, whatever badges we share.
The curve
Who I’m up against isn’t in my industry.
A shared tool would mean nothing if it were Excel. But this one improves as everyone leans on it — so the curve is the competition.
Excel
Everyone has a spreadsheet open, and no one is on anyone else’s curve because of it. Excel doesn’t improve at my job when someone in another field uses it.
The shared engine
Improves release over release. When underwriters, radiologists and recruiters push it harder at screening, the next version is better at screening mine too.
SAME CURVE The finance analyst and I — different fields, one shared engine
NOT ON IT The IT colleague who barely touches the tool — whatever badges we share
“Your competition isn’t in your industry” overstates it, and it’s worth being precise about where. The tool is horizontal — one engine, improving on the same curve for all of us. What stays stubbornly vertical is everything that decides whether its output is safe to trust: validation, risk tolerance, liability. A false positive in a résumé screen costs an awkward phone call; the same miss on a dental X-ray costs an unnecessary surgery, and on a chemical-QA line, a seven-figure shutdown. Same verb, same model, wildly different consequence — and that consequence is the one thing the shared engine can’t standardize for me. That’s the layer where industry stops being a costume and becomes a real container: the tool is everyone’s, but the accountability for what it gets wrong stays local, and mine.
Same verb, different stakes
The tool is horizontal. The accountability stays vertical.
One verb — screening — one model. The cost of a false positive is where industry stops being a costume.
a false positive costs an awkward phone call
the same miss costs an unnecessary surgery
on the shop floor, a seven-figure shutdown
HORIZONTALOne engine, improving on the same curve for all of us.
VERTICALValidation, risk tolerance, liability — what decides if the output is safe to trust.
Someone had said it outright before I ever worked it out. The most radical presenter in the room argued the point directly: stop analyzing jobs and industries, analyze tasks — build the work around who does which verb, human or machine. He’d put it in blunter words than I’d managed, and I’d waved it off as just his field’s angle and moved on.
Some of this the class was assigned to find — “explain how AI will help your industry” pre-loads its own answer. But the prompt only manufactured the agreement, the dutiful “augment, don’t replace.” It never told anyone to reach for Copilot, or to describe screening, or to build an industry talk on the identical piece of software. The consensus was an artifact. The shared engine under it wasn’t.
Artifact vs. engine
The prompt manufactured the agreement — not the engine under it.
“Explain how AI will help your industry” pre-loads its own answer. But the prompt only produced the consensus; it never told anyone what to reach for.
Manufactured by the prompt
The dutiful “augment, don’t replace”
An optimistic verdict on one’s own field
An answer shaped like the question
Not manufactured — the tool’s doing
Everyone reaching for Copilot
Everyone describing screening
Two industry talks on identical software
The most radical presenter said it outright: stop analyzing jobs and industries — analyze tasks. Build the work around who does which verb, human or machine.
What I’m keeping
I’d reached for industry as the thing that explained everyone, and that reach was the exact error the assignment was built to expose. If the costume fooled me with the slides open in front of me, it has surely fooled me in my own career — where the label is more comfortable, and nobody hands the work back with notes.
Negotiating for clarity usually means the work of being understood across an accent. This was the same discipline turned on my own certainty: refusing to accept the word on the title slide as a description of the thing inside. My slide said “IT.” The tool said Copilot, and Copilot has never once cared what industry I think I’m in.
Thanks for reading. If you want one concrete thing to do with it: find the field that leans on your core tool harder than yours does — more volume, higher stakes — and read what they’re fighting about this quarter. Not all of it will carry; an underwriter’s compliance headache may be nothing to me. But they’re pushing the same engine on the same verbs I depend on, so the part that does carry is the closest thing I’ll get to a changelog for my own job — shipped a quarter early, in a costume I’d never have thought to open.
What I’m keeping
Read the field that leans on your tool harder than you do.
One concrete thing to do with all this — the closest I’ll get to a changelog for my own job, shipped a quarter early.
Find the field that pushes your core tool harder than yours does — more volume, higher stakes.
Read what they’re fighting about this quarter — the arguments, the complaints, the edge cases.
Keep the part that carries. Not all will — but they push the same engine on the same verbs, so that part is your job’s changelog, early.
Comments