Task Redefinition
Changing what a system is asked to do so that it becomes tractable, which is how most successful automation actually happened.
Reviewed July 16, 2026Stable
When not to use it
- As a criticism. Redefinition is legitimate engineering and usually the reason a system exists at all.
- To describe all engineering, which would make the term uninformative. The test is whether a specific sub-task or tolerance was changed, not whether the solution differs from the naive one.
- Where the specification genuinely did not move, in which case the capability claim stands as made.
Reach for something else instead
- Capability improvement — meet the original specification, which is what the research frontier attempts and what deployment rarely waits for.
- Human-machine division — assign the non-negotiable part to a person, which is teleoperation and shared autonomy.
- -
Read more on the blog
- Ten domains, and the specification moved in every oneTerritory 7 closes. Physical automation succeeded wherever the task could be changed, and the one place it has not is where nothing was negotiable.
- 0.03% of construction spend, so the work moved indoorsOn a building site nothing about the task is negotiable. The industry's answer was not a better robot but a different location.
- 32 million vacuums, and 38% on household tasksThe most-deployed consumer robot succeeded by accepting a worse result more often. The home is the one place where the other ways of making a task tractable are unavailable.
- The best-documented humanoid deployment is seven unitsCirculated figures run to tens of thousands. Company statements run to hundreds, and one chief executive said his robots are not in material use.
Further reading
- International Federation of Robotics, World Robotics 2025 — the installed base that environment engineering produced.
- Kusano et al. (2025), Comparison of Waymo Rider-Only crash rates by crash type to human benchmarks at 56.7 million miles — domain narrowing, and a benchmark correctly adjusted to it.
Primary sources, listed so you can check the claims on this page rather than take them on trust.
Where people go wrong
- Reading a deployment as evidence for the original, harder task.
- Treating disclosure of redefinition as a weakness rather than as the information needed to interpret the result.
- Assuming a redefinition available in one domain transfers to another; breeding a crop has no analogue in most settings.
At a glance
FieldFoundations
Five formsenvironment engineering, domain narrowing, target standardisation, sub-task deletion, tolerance widening
Key questionwhich parts of the task are negotiable
Evaluation consequenceresults do not transfer to the original task
Hard casegeneral-purpose manipulation, where nothing is negotiable
DifficultyIntermediate
Where this sits
A starting point. Nothing needs to come before it.
0Levelstarting point
0Needs firstnothing
0Opens upnothing further
1Areastays here
Computed from the prerequisite graph, not assigned. How this works