When machines do the work, people are left setting the limits.
Every instrument we have for setting limits on work — job design, code review, human oversight, privacy law — assumes a person is doing it. These papers ask what happens to each one when a system does the work instead.
Four papers, so far.
The Mixed Workforce Company
An interactive model for splitting work, authority, and accountability between people and AI.
Interactive paper · 12 min readDoes Code Still Need to Be Maintainable?
A blinded replication comparing clean and deliberately sloppy code across repeated maintenance tasks. The updated study examines task success, retries, and cost, with explicit limits on what the results establish.
Interactive benchmark · 6 min readThe Human Exclusion Threshold
An interactive inquiry into the point where operational speed, complexity, and autonomy make meaningful human intervention impossible—and move human authority upstream.
Interactive model · Theory + labThe Right to Be Forgotten Needs Rewriting
Erasure works when a controller can find the thing it is deleting. In a trained model one person's data shifts millions of parameters without occupying a location, so the remedy has no target. This paper maps six layers at which deletion can happen, tests whether current unlearning survives adversarial probing and routine 4-bit compression, sets four European regulators' incompatible answers against each other, and argues for replacing erasure with a use restriction plus a process-based standard the controller has to prove.
Interactive paper · 15 min read