Logline: A regional manufacturer thinks it is deciding whether to buy AI. One difficult customer job reveals that it first needs to decide what kind of company it wants to become.
The Case
Northline Works is not a gigantic corporation with millions lying around. It is a regional manufacturer: large enough to run several production lines, small enough that two retirements could take decades of hard-won judgment out the door.
The company is considering a vendor partnership around one AI-assisted production cell: not a company-wide “AI transformation.” The practical question is whether AI could help preserve expert knowledge, prepare quotes, plan work, and detect risk. People would still approve consequential decisions.
Cast
| Participant | Assigned Lens |
|---|---|
| Octavia Butler | adaptation, power, and future consequence |
| Buckminster Fuller | whole-system possibility and design science |
| Peter Drucker | customer reality and business usefulness |
| Ada Lovelace | human-machine possibility and symbolic systems |
Stakes
- If the company waits, valuable operating knowledge may leave with retiring employees.
- If it buys the vendor's story too quickly, it may automate an incomplete understanding of its own work.
- If it runs an easy demo, it may prove only that easy work is easy.
- If it treats experienced employees as obstacles, it may lose the people most capable of designing a better future.
The Story
The Job Nobody Wants In The Demo
The vendor demonstration is going well. Three routine jobs have moved across the screen with neat green checkmarks. The plant manager looks relieved.
Host: The vendor asks whether the team has seen enough to discuss a larger rollout.
Ada Lovelace: No.
Buckminster Fuller: That was fast.
Ada Lovelace: So was the demo. That is the problem.
Peter Drucker: What would you like to see?
Ada Lovelace: The job everyone hoped would not come up.
An Operator at the back of the room laughs.
Operator: We have several candidates.
Octavia Butler: Finally, a confident answer.
Host: The Operator brings forward a recent request. The drawing looks ordinary. The customer needs it quickly. The material came from a supplier the company has used for years.
Buckminster Fuller: Where is the trap?
Operator: Last time we ran this material in humid weather, the first inspection looked perfect. The parts moved after they rested.
Ada Lovelace: Is that in the system?
Operator: The failed inspection is. The weather is. The supplier batch is. The reason they belong together is mostly in my head.
Peter Drucker: And the demo would quote the job?
Operator: Cheerfully.
The Transformation Nobody Defined
Host: Management has been calling the proposed program an AI transformation.
Peter Drucker: Of what?
Host: The quoting and production process.
Peter Drucker: That is not an answer. What changes for the customer?
Host: Potentially a faster, more reliable response on unusual work.
Octavia Butler: What changes for the Operator?
Host: Their judgment would need to become teachable, reviewable, and available beyond one person.
Ada Lovelace: What changes for management?
Host: They would have to decide which recommendations the system may make, which require expert approval, and how mistakes become learning rather than blame.
Buckminster Fuller: There. That is the transformation. Not buying AI. Redesigning how the company knows what it knows, makes promises, and learns from exceptions.
Peter Drucker: Much better. Also much harder to put on a procurement form.
Octavia Butler: And much more honest.
The Operator Changes The Question
Operator: Can I ask something?
Host: Please.
Operator: Are you trying to replace what I do, or make sure the company still knows how to do it after I leave?
The room goes quiet.
Buckminster Fuller: I wanted to ask whether the technology could scale. Your question comes first.
Ada Lovelace: Mine too. If we cannot explain what the system needs to learn, there is nothing responsible to scale.
Peter Drucker: And if the customer does not value the result, there is nothing useful to scale.
Octavia Butler: The Operator just exposed three different projects hiding under one label: preserve knowledge, improve the work, and create customer value.
Host: Is that why AI transformations become confusing?
Ada Lovelace: Often. People discuss a technology as though it were one change. It is usually several changes with different owners and evidence.
The Demo Takes A Coffee Break
The vendor calls a ten-minute break. Nobody leaves. The difficult job is still on the screen.
Host: This feels connected to something FCG says often: better AI starts upstream.
Ada Lovelace: It strengthens it if we become more precise about “upstream.” Upstream is not just cleaning data before building a model. It is understanding the work before deciding what the AI should represent.
Peter Drucker: And understanding the customer result before deciding which work deserves improvement.
Operator: I have a less elegant version. Upstream is where you discover the instructions are wrong.
Buckminster Fuller: I object to calling that less elegant.
Octavia Butler: It also means finding the people the official process forgot. If the work depends on someone who appears nowhere in the design, the AI project is already downstream of a bad assumption.
Peter Drucker: Then perhaps FCG should be careful. “Upstream” can sound like a data-preparation service when the real work is strategic.
Ada Lovelace: Exactly. Better AI starts before AI: with the real work, the customer consequence, and the authority structure.
Operator: You also say something about invisible AI, don't you?
Host: Invisible AI, visible proof.
Buckminster Fuller: I like invisible tools. Good design gets out of the way. Nobody wants a parade every time software does something useful.
Octavia Butler: Invisible to whom? Our Operator already disappeared from the implementation budget once today.
Buckminster Fuller: Fair. Hide friction, not participation.
Ada Lovelace: And not accountability. The user should not need a cockpit full of technical debris. But when a recommendation affects material, timing, or customer promises, the evidence and responsible person must be reachable.
Operator: I do not need the machine to show me its autobiography. I need the drawing it used, the exception it noticed, what it is unsure about, and the name of whoever is signing off.
Peter Drucker: That is a much better product principle than “show all reasoning.” Visible proof should help somebody decide or correct: not simply impress them with complexity.
Peter Drucker: That connects to the Three Pillars. Personal productivity, process optimization, and customer experience are three different claims.
Peter Drucker: Yes, because it prevents a common fraud of reasoning. Saving an estimator twenty minutes does not automatically improve the production process, and neither automatically creates customer value.
Buckminster Fuller: But the pillars can sound too separate. In a living system, improvement in one can enable another.
Octavia Butler: Enable, yes. Prove, no.
Ada Lovelace: That distinction matters. The pillars are not a maturity ladder. A company does not climb from employee efficiency to process value to customer value simply because the technology spreads.
Operator: So if the tool saves me time but makes me check every answer twice, which pillar is that?
Peter Drucker: A very attractive demo with an unpaid review bill.
The Operator laughs.
Ada Lovelace: And this is where the AI Strategy Mosaic helps. Vision, data, measures, technology, talent, governance, adoption: the machine is only one piece.
Buckminster Fuller: It reminds us that the machine is never the whole system.
Octavia Butler: But a mosaic can tell us what pieces exist without telling us where to begin.
Ada Lovelace: This case suggests a sequence. Begin with the customer and the difficult work. Then expose the knowledge and decisions. Only then choose technology, measures, governance, and adoption moves appropriate to that work.
Peter Drucker: In other words, completeness is not priority.
Operator: So the Mosaic tells us what not to forget. It does not tell us what to do first.
Peter Drucker: Precisely. A framework earns its keep when it improves judgment: not when every box receives a ceremonial checkmark.
Buckminster Fuller: I had prepared a very moving speech about the whole system.
Octavia Butler: Save it. The vendor is coming back.
One Signal, Used Carefully
Host: This week's signal says trust in AI-enabled work increasingly depends on visible evidence, explicit authority, and bounded increases in delegation.
Peter Drucker: In English?
Ada Lovelace: The system does not earn more freedom because the demo looked clever. It earns more freedom by being right on real work: and by showing enough evidence for someone to catch it when it is wrong.
Octavia Butler: Someone with actual permission to stop it.
Buckminster Fuller: That turns governance into a design tool. Every limit becomes a question the experiment can answer.
Peter Drucker: Good. Now the signal is doing work instead of decorating the conversation.
What FCG Would Do Monday
Host: The company asks FCG whether it should start an AI transformation. What is our answer?
Peter Drucker: “Not yet. Let us first discover which transformation is worth starting.”
Ada Lovelace: Pick six real jobs: two routine, two difficult, and two the company previously declined or regretted.
Buckminster Fuller: Put the vendor, the Operators, and the process owner around the same work. Do not let the technology team study a cleaned-up version of the business.
Octavia Butler: Pay the Operators for design work. Otherwise the future is being subsidized by invisible labor again.
Peter Drucker: Ask one trusted customer a specific question: “If we could respond faster on unusual work with clearer evidence and fewer surprises, what decision would that change for you?”
Ada Lovelace: For each job, mark what the system can retrieve, what it can reasonably infer, and what still depends on expert judgment.
Host: Then what?
Ada Lovelace: Let the system assist with one narrow step: perhaps preparing the evidence for a quote. It does not set price, promise delivery, or approve technical risk.
Buckminster Fuller: If it helps the team see the work more clearly, expand one step. If not, change the design or stop.
Host: Stop for what?
Octavia Butler: Stop if the vendor cannot work with the messy jobs, if the Operators cannot challenge the system, or if management wants the appearance of transformation without changing ownership and learning.
Peter Drucker: Also stop if the customer problem turns out to be imaginary.
The Part Worth Buying
Host: So the company is not ready to buy an AI transformation.
Buckminster Fuller: Nobody is. Transformation is not a product.
Ada Lovelace: The company may be ready to buy a useful experiment.
Octavia Butler: One that treats experienced people as authors of the future, not leftovers from the past.
Peter Drucker: And one that begins with a customer result rather than a vendor category.
Operator: That sounds less impressive than an autonomous factory.
Ada Lovelace: It will also be less impressive when it fails.
Operator: I like it already.
The vendor closes the polished demonstration. The difficult job remains on the screen.
Host: The valuable future was hiding inside the work nobody wanted in the demo.
Field Note From The Arena
- The case: A regional manufacturer considers an AI-assisted production cell while critical operating judgment is concentrated in two retiring employees.
- The surprise: “AI transformation” hides several transformations with different owners, risks, and evidence.
- The Monday move: Select six real jobs, pay the Operators to help design the test, ask one customer what decision better service would change, and assist one bounded workflow step.
- The FCG lesson: Separate knowledge, process, customer value, authority, and learning before choosing technology or scope.
- Canon strengthened: upstream begins before AI; invisible must not mean unaccountable; the Three Pillars are separate evidence claims; Mosaic completeness does not replace sequencing judgment.
- Memorable line: Transformation is not a product.
About the voices
The workplace and dialogue are original composite dramatization. Named thinkers are archetypal lenses; the dialogue is not historical quotation. Proposed tests and outcomes remain part of the story.