The task-based fallacy.
AI-jobs analysis treats a job as a fixed list of tasks AI either does or doesn't. The data shows a job is a bundle, held together by a constraint. AI restructures the bundle, not the task list.
What the fallacy claims.
The dominant frame for AI's labor impact runs like this: a job is made of a list of tasks. AI does some of the tasks. The job either survives (the human keeps the other tasks) or it doesn't (AI does them all). The "what does AI automate?" question is the characteristic move.
This is the task-based fallacy. It treats a role as decomposable into its task inventory, with no remainder. Drop one task, the role shrinks by that much. Drop all of them, the role is gone.
A role is a bundle, not a list.
A role is a position in a system of work - a bundle of components held together by a binding constraint. The constraint is what makes the components live in one job rather than several.
When AI enters the system, the components don't disappear one by one. The constraint that bound them together loses force, and the bundle scatters. The components persist; the role that held them in one seat does not.
Three things can happen to any individual component:
- The tool absorbs it. AI does the task directly. This is the only case the task-based fallacy gets right.
- The skill premium gets eaten. A less-skilled worker, with AI, now performs the expert task. The component moves to a different worker, not to AI.
- The bundle unbundles. The task wasn't touched by AI at all. But the bundle that held it broke, and the work moved to a different role entirely. The fallacy has no frame for this third mechanism.
The third mechanism is where the fallacy fails hardest. Looking only at "which tasks did AI automate?" misses every component that relocated to brand teams, project managers, platform terms-of-service, or institutional accountability seats - all of which appear in the translator data.
The diagnostic.
An AI-jobs claim is leaning on the task-based fallacy when it:
- Lists tasks and asks "which can AI do?" without asking what binds them together as a role.
- Concludes the job survives because some tasks aren't automatable yet.
- Treats workers as task-bundles to be unwound piece by piece, rather than as occupants of a position in a value chain.
- Ignores where the components go when they leave the role (other roles, platforms, contracts, institutions).
The corrective is to ask: what constraint binds this role? Where is that constraint going? And where are the components moving to? Those are bundle questions, not task questions.
The translator data.
In the translator case study, the translator's old bundle had ten components. Only four were absorbed by AI directly. Five were unbundled and relocated to other roles - cultural mediation to brand teams, client scoping to project managers, certification to institutional roles. A pure "which-tasks-did-AI-do?" reading misses 50% of the bundle's dissolution.