Definition
The Fade — also called Uneven Disappearance — is the central thesis of Greg Kihlström‘s book Stop Saying “AI”. It holds that the word “AI” is on its way out of everyday language, and that the way it leaves matters more than the fact that it does.
The historical model is electricity, not the internet. We once said electric light, electric motor, electric refrigerator. The qualifier dropped off the moment electricity stopped being remarkable and became an assumed property of the things around it. “AI-powered” is that qualifier today — AI-powered search, AI-powered writing, AI-powered analytics — and it’s on the same path. When intelligence becomes the assumed substrate, the prefix becomes redundant and disappears. You won’t have an AI assistant. You’ll have an assistant, and the intelligence will be presumed.
The important part is what the analogy gets wrong if taken too far. Electricity became invisible because it became reliable and standardized. A wall socket delivers the same current every time, and when it fails it fails obviously. AI is probabilistic and fallible. It doesn’t return the same answer twice, and it fails quietly — a confident wrong answer looks exactly like a confident right one. So the disappearance is uneven. AI fades into the background for low-stakes, ambient uses, and it stays explicitly named and explicitly watched wherever being wrong carries cost.
The honest formulation isn’t “AI becomes invisible.” It’s “AI becomes invisible everywhere except where its fallibility carries consequences.” That caveat is what keeps the thesis defensible, and it’s the hinge the rest of the AI Capability Framework turns on.
How It Relates to Marketing
The Fade reframes a comfortable statement — “we’re an AI-powered company now” — as a strategic risk. The things people stop noticing are the things they stop managing. A capability that disappears into every workflow is one nobody owns, budgets, or audits. In marketing, that describes a large and growing share of the stack.
For marketing leaders, the practical consequence is a sorting job. Some AI uses should fade, and letting them do so is correct. Others must be held in the light. The Fade gives that sorting a principle:
- Uses that can safely fade — subject-line variants, send-time optimization, ticket routing, draft assistance. If one is wrong, the cost is a shrug, and someone would notice quickly.
- Uses that must stay named — a claim denial, a credit decision, a price quote, anything a regulator or a furious customer will demand you explain. A quiet failure here costs money, a customer, or an account of yourself you owe someone angry.
The line between the two piles isn’t about how impressive the technology is. It’s about what being wrong costs, and whether anyone would notice in time.
Why the Fade Is Uneven
Every ambient technology had to earn its invisibility through infrastructure that made disappearing safe. Electricity needed standardized voltages, grounded wiring, fuses, and safety codes before anyone could stop thinking about it. The web needed certificates and a trust apparatus before commerce could treat it as plumbing.
AI’s equivalent is a trust and verification layer — provenance, audit trails for autonomous actions, calibrated confidence, and clear lines of authority for what a system may do and on whose behalf. For a probabilistic technology this isn’t optional polish. It’s the load-bearing condition that makes the fade possible at all. Electricity got fuses before it got to be invisible. AI needs its equivalent before “AI-powered” can safely drop the prefix — and in the places where that infrastructure isn’t yet trustworthy, the capability stays named by necessity.
How to Apply the Concept
- Inventory what you already call “AI.” List every feature and vendor currently filed under the umbrella term.
- Ask the cost-of-wrong question for each. If it produces a wrong output, does anyone notice immediately, or does it fail quietly? Does a wrong output cost money, a customer, or a regulator’s attention?
- Sort into fade and stay-named. Loud, cheap failures lean toward fade. Silent, expensive failures stay named and need a trail.
- Stop managing the fade pile as projects. Let low-stakes capabilities become assumed plumbing. Naming them in every deck is wasted attention.
- Invest oversight where it counts. Concentrate governance, verification, and human review on the stay-named pile, where being forgotten is the failure.
Comparison to Similar Concepts
| Concept | Focus | Relationship to the Fade |
|---|---|---|
| The Fade (Uneven Disappearance) | A technology’s name dropping as it becomes assumed, unevenly | The core thesis |
| General-purpose technology | Technologies that diffuse across the whole economy | Explains why AI diffuses broadly, but not the uneven part |
| Wardley Mapping | Components evolving toward commodity | Shares the drift-to-infrastructure idea; lacks the fallibility caveat |
| Commoditization | Products losing differentiation over time | Related, but about price and margin rather than language and management |
The Fade is distinguished from ordinary commoditization by the fallibility caveat. A commodity becomes invisible because it becomes uniform and safe. AI can’t fully do that, because it can be confidently wrong, which is why the disappearance splits into two piles instead of completing.
Best Practices
- Treat disappearance as a decision, not a milestone. Choose what fades. Don’t let it happen by inattention.
- Keep the caveat attached. Any time the electricity analogy is used, carry the “except where being wrong is costly” clause. Without it the thesis flatters teams into ignoring their highest-stakes uses.
- Watch for silent failure specifically. The dangerous pile isn’t the loud errors. It’s the quiet ones that look like success.
- Re-sort periodically. A use that was low-stakes can move as it scales or touches a decision it didn’t before.
Future Trends
- The prefix erodes faster in consumer products than in regulated ones. Expect “AI-powered” to vanish from everyday tools while persisting in finance, healthcare, and legal contexts, exactly along the cost-of-wrong line.
- Trust infrastructure becomes the gating factor. The pace of the fade in any given domain will track how mature its verification and provenance tooling is.
- Naming becomes a governance act. As the word fades, deliberately keeping a capability named will function as a control in its own right.
FAQs
1. Where does the term come from? The Fade is the central argument of Stop Saying “AI” by Greg Kihlström.
2. Why electricity and not the internet? The internet is a place you go — you “get online,” you “look it up” — so it kept its name. Electricity is a property things have, so its name fell away. AI is becoming a property, not a place, which is why its label will erode like “electric” did.
3. Does the Fade mean AI stops mattering? The opposite. Invisible isn’t unimportant — it’s the reverse. The capabilities you stop noticing are the ones you stop managing, which makes the disappearance a strategic event.
4. What makes the disappearance uneven? AI is probabilistic and fails quietly. It fades where being wrong is cheap and obvious, and stays named where being wrong is costly and silent.
5. Is the Fade the same as commoditization? Related but not identical. Commoditization is about price and uniformity. The Fade is about language, attention, and management, and it never fully completes because AI can be confidently wrong.
Related Terms
- The AI Capability Framework (Master Stack)
- Place vs. Property
- Cost-of-Wrong
- The Fade / Stay-Named Line
- The Spine
- The Trust / Verification Layer
- Artificial Intelligence (AI)
- Sovereign AI
- Environmental Implications of AI Usage
- Wardley Mapping
Sources
- Kihlström, Greg. Stop Saying “AI”. https://amzn.to/4wilWcA
- Greg Kihlström — official site. https://www.gregkihlstrom.com
- David, Paul A. “The Dynamo and the Computer: An Historical Perspective on the Modern Productivity Paradox.” American Economic Review, vol. 80, no. 2, May 1990.