This is part of a series. The Gutenberg/Semantic model describes physical versus logical layers. The Def-Use split describes how authors and users inhabit different semantic models. CarPlay and Nokia showed the certification tribe. This post shows three more versions of the same cycle — one app, one chipmaker, one company that kept escaping.

1. Sonos: Breaking the Semantic Layer While the Gutenberg Layer Was Perfect

Sonos built some of the best consumer audio hardware ever made. The speakers, the amplifiers, the acoustic engineering, the multi-room synchronisation — all excellent. Pure Gutenberg: physical devices that move air in precisely controlled ways. Users trusted Sonos hardware deeply, paid premium prices for it, and built whole-home systems around it.

In May 2024 Sonos shipped a ground-up app rewrite and destroyed that trust in a single release.

The redesign was driven by internal architectural goals — a new platform to support future hardware, a cleaner codebase, a new design language. All legitimate Def concerns. But Sonos shipped the Def before the Use was ready. Core features that had worked for years were missing or broken at launch:

  • Alarm management gone
  • Local music library support broken
  • Accessibility features removed
  • Queue management degraded
  • Sleep timer missing

Not bugs in new features. Deliberate removal of existing Use to ship on schedule. The internal semantic model — new architecture is better — overrode the external Use signal — users need these features to listen to music. The Gutenberg layer (the hardware) remained perfect. The semantic layer (the app) was broken. Expensive, well-engineered speakers made frustrating by a broken interface. The iceberg upside down.

The tribal defence was textbook. CEO Patrick Spence’s initial response was essentially “you’ll get used to it” — the classic Def-push posture. The Def is correct. The Use must adapt. It took months of sustained user fury, App Store reviews cratering to 1 star, and measurable hardware sales decline before Sonos publicly apologised. Spence resigned in early 2025.

The sharpest point is that Sonos users were not asking for new features. They were asking to keep the ones they had. This is the most basic possible Use signal — do not take away what works — and Sonos filtered it out entirely. The Def was so consuming internally that existing Use became invisible.

The premium audio community and the custom installation market (professional AV installers who certified and sold Sonos systems) defected immediately and loudly. The tribe of loyal users became the amplifier of the Use signal rather than its suppressor — the opposite of the certification-as-defence pattern. When your most invested users are also your most vocal critics, the feedback loop is working. The tragedy is that Sonos heard it too late.

2. Intel: The Fab Tribe

Intel’s decline from semiconductor dominance is the manufacturing tribe defending their fabs past the point where the fabs could defend themselves.

Intel’s integrated model — design the chip, manufacture the chip, in your own fabs, at the leading edge — was a source of genuine competitive advantage for decades. The fab was not just a Gutenberg asset. It was tribal identity. “Intel Inside” meant Intel-designed and Intel-manufactured. The two were inseparable in the tribal Def.

TSMC built a different model: pure Gutenberg. Manufacture chips. Anyone’s chips. No tribal identity in the design, no competitive interest in what the chip does semantically. Just physics, process nodes, and yield rates. TSMC’s semantic layer is the foundry service. The Gutenberg layer is the fab itself — and TSMC invested in that Gutenberg layer with a discipline and capital intensity that Intel, managing both design and manufacturing simultaneously, could not match.

The Use signal arrived through Apple. Apple had been designing its own chips since the A4 in 2010, fabbed at Samsung and then TSMC. The M1 in 2020 demonstrated that a chip designed with specific Use cases in mind — the Mac’s actual workloads — and fabbed at TSMC’s leading process node outperformed Intel’s best in performance per watt by a significant margin. The Def (Intel’s integrated model) met the Use (what Apple’s customers actually needed from a laptop chip) and lost.

Intel’s tribal response was to defend the integrated model. The fabs were identity. Spinning them off, as GlobalFoundries had done from AMD, felt like defeat. Intel Foundry Services was announced as a way to have both — keep the fabs, open them to external customers, compete with TSMC — but the tribe’s identity made it difficult to fully commit to either model. You cannot be both the best chip designer and the best chip manufacturer when TSMC has spent thirty years doing nothing but manufacturing.

The certification parallel: Intel’s tight integration with Microsoft (Wintel) was its own certification tribe. Windows was optimised for x86. x86 was Intel. The ecosystem lock-in was mutual and reinforcing — until Apple demonstrated that the semantic layer (macOS, iOS, the app ecosystem) could be cleanly separated from the Gutenberg layer (the instruction set architecture) with a translator in between.

3. Apple’s Architecture Transitions: The Cleanest Gutenberg/Semantic Separation

Apple has switched CPU architectures three times in the Mac’s history. Each transition is a masterclass in Gutenberg/Semantic separation:

  • 1994: Motorola 68k → PowerPC
  • 2006: PowerPC → Intel x86
  • 2020: Intel x86 → Apple Silicon (ARM)

In each case the semantic layer — macOS, the application ecosystem, the user’s software investment — had to survive a complete change in the Gutenberg layer (the instruction set, the physical chip, the hardware platform). Apple’s solution was Rosetta: a binary translation layer that runs old Gutenberg instructions on new Gutenberg hardware by translating them in real time.

Rosetta is elegant precisely because of where it sits in the stack. It is a Gutenberg-to-Gutenberg translator — it takes x86 bytecode (Gutenberg addresses and operations for the old hardware) and translates them to ARM bytecode (Gutenberg addresses and operations for the new hardware) without the semantic layer ever knowing the transition happened. The user’s application continues to behave semantically identically. The Gutenberg substrate has changed completely underneath.

This is the resolver pattern applied to CPU architecture. Rosetta is DNS for instruction sets: the semantic name (your application, your data, your workflow) resolves to a different Gutenberg address (a different instruction encoding) transparently. The semantic layer is stable. The Gutenberg layer is swappable.

The x86 → Apple Silicon transition is the sharpest example because the motivation was explicitly Gutenberg-layer improvement. Intel’s fabs had fallen behind TSMC’s process nodes. Apple was paying for Gutenberg (manufacturing physics) that was no longer the best available. The semantic layer — macOS, the developer ecosystem, the user’s investment — did not need to change. Only the Gutenberg substrate needed to improve. Rosetta made the transition possible in a single OS release, with most users barely noticing during the overlap period.

Apple’s willingness to make this transition — three times — reflects a consistent architectural posture: the semantic layer must not be coupled to the Gutenberg layer. The application ecosystem, the developer tools, the user experience are the product. The chip is infrastructure. When better infrastructure is available, you switch. The semantic layer travels with you.

Intel’s integrated model made this posture impossible internally. When the design and the fab are the same tribe, changing the fab means changing the tribe’s identity. Apple had no such constraint — they designed the chips but never fabbed them. The Gutenberg layer was always someone else’s problem, which meant it was always replaceable.

4. The Common Thread: Tribal Identity in the Gutenberg Layer

Sonos, Intel, and Nokia share a specific failure mode: the tribe formed around the Gutenberg layer and then claimed ownership of the semantic layer too.

  • Sonos’s Gutenberg layer is the hardware. Their tribal identity extended to owning the app — the semantic interface — even when the hardware business would have been better served by letting the semantic layer evolve in response to Use.
  • Intel’s Gutenberg layer is the fab. Their tribal identity extended to the chip design and the Wintel ecosystem — semantic claims built on top of a Gutenberg asset.
  • Nokia’s Gutenberg layer was the handset manufacturing and carrier relationships. Their tribal identity extended to the OS and the user experience — semantic territory that Symbian couldn’t defend when the Use signal moved.

Apple’s consistent advantage is that they form the tribe around the semantic layer — the user experience, the ecosystem, the software — and treat the Gutenberg layer as replaceable infrastructure. When the Gutenberg layer (Motorola, PowerPC, Intel, eventually TSMC’s nodes) no longer serves the semantic layer best, they switch. The tribe’s identity is not at risk because the tribe is defined by the semantic product, not the physical substrate.

This is the pace layering principle correctly applied: the semantic layer (your product, your user relationship) should be more stable than the Gutenberg layer (your manufacturing process, your chip vendor, your hardware platform). When the Gutenberg layer changes faster than the semantic layer can absorb, you need a translator — Rosetta, or an emulator, or a compatibility layer. When the Gutenberg layer stops changing fast enough, you switch substrates and let the translator handle the transition.

The weak link willing to learn treats the Gutenberg layer as a variable and the semantic contract with the user as the constant. The tribe gets it backwards: the Gutenberg asset (the fab, the handset, the speaker hardware) becomes the constant, and the user’s semantic experience becomes the variable that must adapt.

5. Rosetta as the Model

Rosetta deserves its own moment because it is the most elegant engineering solution in this series.

Every architecture transition creates a period where the new Gutenberg layer exists but the semantic ecosystem has not yet been rewritten for it. Native apps for the new architecture are faster. Old apps still need to run. The naive solution is to maintain two separate ecosystems — the old Gutenberg world and the new one — in parallel until the old one dies. This is painful, expensive, and creates a long tail of legacy applications that never get rewritten.

Rosetta’s solution is to make the Gutenberg translation transparent and fast enough that the semantic layer does not need to know it is happening. Old x86 binaries run on ARM hardware at acceptable (and in Apple Silicon’s case, often surprisingly competitive) performance. Developers rewrite their apps for native ARM on their own schedule, pulled by performance gains rather than pushed by compatibility deadlines. The transition happens gradually, Use-signal-driven, without a hard cutoff.

This is Build-Measure-Learn applied to platform migration. Ship the new Gutenberg layer with a translator. Measure which semantic applications get rewritten natively (the ones users care about most). Learn which parts of the old ecosystem can be retired. The translator buys time for the Use signal to determine what actually needs to migrate.

The alternative — hard cutoffs, forced migrations, “your old software will not run” — is the train. Rosetta is the open road with a GPS that quietly reroutes around the roadwork.

6. The Pattern Holds

Sonos, Intel, Apple’s transitions, Nokia, CarPlay, MDI, Eclipse, IE — the same seven-step cycle:

  1. A real Gutenberg asset produces a legitimate competitive advantage
  2. The tribe forms around the Gutenberg asset
  3. The tribe extends its Def to claim the semantic layer too
  4. The Gutenberg asset stops being best-in-class (fabs fall behind, hardware matures, screens get bigger)
  5. The Use signal arrives — users want the semantic layer to evolve independently
  6. The tribe filters it out — the Gutenberg asset and the semantic Def are the same identity
  7. A new entrant with no tribal investment separates the layers cleanly and wins

Apple escapes the cycle by repeating step 4-7 on themselves before a competitor does it to them. The tribe is defined by the semantic layer, so switching the Gutenberg substrate is not an identity threat — it is just good engineering.

The Def that survives is the one that knows which layer it actually lives in.