Three mechanisms of bundle breakage.
A role bundle can break three ways under AI pressure. The automation-augmentation binary names two. The third - the task wasn't touched by AI, but the bundle that held it broke - is the one routinely misread as "the work is gone."
Three different mechanisms.
A role bundle is a set of components held together by a binding constraint. When AI enters the system, components leave the bundle for one of three reasons. Each mechanism leaves a different fingerprint in the labor data; conflating them obscures what's actually happening.
- Tool absorbed. AI does the task directly. The work happens inside the model; no human performs it any longer. This is the case the field calls "automation." Both the automation-augmentation binary and the task-based fallacy capture this mechanism correctly.
- Skill premium eaten. AI augments a less-skilled worker to perform the expert task. The task itself is still done by a human, but the human's defensible skill edge has been compressed. Pay falls; the role shifts down the wage ladder. This is the case the field calls "augmentation."
- Unbundled & relocated. The task was not touched by AI. The bundle that held it broke. The work moved to a different role entirely - to a different department, a different contract, a different institution. The task persists; the original bundle no longer reaches it.
Why mechanism 3 is invisible to the binary.
The automation-augmentation binary asks two questions: "did AI do this?" and "did AI help someone less skilled do this?" When the answer to both is "no," the binary concludes "this work survives unchanged."
But the work can persist while its bundle home dissolves. Cultural mediation in translation didn't get automated. AI doesn't do it. It also didn't get augmented down the wage ladder. Instead, the task left the translator's seat and moved to brand-marketing teams and in-country reviewers - different people, different contracts, different reporting lines.
From the binary's perspective, this looks like "the task is gone." From the bundle's perspective, the task is fine - it's just not in the translator's bundle anymore. The mechanism is real and structural, but the binary has no language for it.
Diagnosing the mechanism.
For each component of a role bundle, ask:
- Is the task being done by AI inside the model? → tool absorbed.
- Is the task being done by a less-skilled human with AI assistance? → skill premium eaten.
- Is the task still being done by humans, but not by this role? → unbundled and relocated.
The diagnostic is whether the task is happening at all, then whether the role bundle in question is the one performing it. Mechanism 3 is the case where the answer is "task happens; not here."
The translator breakdown.
Across the 10 components of the canonical translator bundle, the breakdown is 4 tool-absorbed / 1 skill-premium-eaten / 5 unbundled-and-relocated. Half the bundle fits the third mechanism - the one the binary misses. The full breakdown is shown in the three-mechanism scoreboard on the page.