As the cost of reliable machine work falls by 90%, which digital and physical tasks become economical first?

Evidence(8)

1. 

   Routine tasks are the initial automation candidates

   [8]

   4 Jan 2016

   The St. Louis Fed reported that employment in routine occupations had been mostly stagnant while nonroutine cognitive and manual employment had increased. That pattern is consistent with repetitive, codifiable tasks becoming economically vulnerable before ambiguous judgment and exception-heavy work.

   1

   This establishes the baseline labor-market pattern: routine work has weaker employment growth than nonroutine work. For the 90%-cost-decline scenario, it supports screening at the task level and prioritizing work that is repetitive and rule-like, while avoiding the assumption that entire occupations disappear at once.

   See how this bears on the question
2. 

   Automation is a task-allocation problem, not an occupation-level switch

   [9]

   2019

   Acemoglu and Restrepo’s framework describes automation as capital taking over tasks previously performed by labor, while new tasks can reinstate labor. Applied to a 90% fall in reliable machine-work cost, the earliest candidates are tasks with clear boundaries, measurable outputs, and limited exception handling—not necessarily whole jobs.

   See how this bears on the question
3. 

   Reliability and recovery costs remain part of the price

   [7]

   11 Oct 2023

   ABB reported that the typical industrial business experiences unplanned outages costing close to $125,000 per hour. This makes machine tending, inspection, packaging, and other repeatable physical tasks more attractive when failures are detectable and recoverable, while safety-critical or highly disruptive tasks remain later candidates.

   See how this bears on the question
4. 

   Computer-vision exposure does not automatically imply economic automation

   [3]

   8 Mar 2024

   The MIT study found that only 23% of worker wages paid for vision tasks were economically attractive to automate at then-current costs, even though 36% of U.S. nonfarm jobs had at least one task exposed to computer vision. The result supports an ordering in which structured inspection and monitoring cross before ambiguous visual judgment.

   See how this bears on the question
5. 

   CAM and toolpath generation become early digital-physical candidates

   [10]

   19 Dec 2024

   Autodesk described AI-assisted CAM programming that generates optimized toolpaths in minutes and removes repetitive programming work. This supports placing template code, tests, and CAM instructions ahead of bespoke software, process engineering, and unstructured design in the economic sequence.

   See how this bears on the question
6. 

   Physical production crosses the threshold when setup costs can be amortized

   [11]

   31 Mar 2026

   Fictiv’s comparison notes that injection molding has high upfront tooling costs but lower unit costs as volume increases. The implication for a 90% decline in reliable machine-work cost is that standardized, high-volume physical tasks become economical before bespoke, low-volume work whose setup, fixturing, and programming costs dominate.

   See how this bears on the question
7. 

   Manufacturing AI favors measurable use cases but remains integration-constrained

   [12]

   13 May 2026

   NIST’s manufacturing guidance describes AI-enabled cameras for product inspection, safety monitoring, and inventory tracking, while identifying data quality, high initial costs, workforce readiness, cybersecurity, and legacy-system integration as barriers. These constraints explain why monitoring, inspection, routing, and machine tending precede ambiguous or safety-critical automation.

   See how this bears on the question
8. 

   Resulting ordering under a 90% decline in reliable machine-work cost

   [8][9][7][3][10][11][12]

   Date unavailable

   Digital first: record extraction and routing, reconciliation, monitoring, standardized inspection, template code and tests, and CAM/toolpath generation.

   Physical next: machine tending, pick-and-place, packaging and palletizing, repeatable inspection, and standardized CNC or other high-volume production.

   Later: bespoke design, ambiguous judgment, negotiation, unstructured physical work, and safety-critical tasks where setup, integration, exception handling, liability, or failure-recovery costs remain high.

   See how this bears on the question