Short-Term Gain, Long-Term Fragility: AI Labor Substitution and the Erosion of Sustainable Capability

Short-Term Gain, Long-Term Fragility: AI Labor Substitution and the Erosion of Sustainable Capability

Source: cs.CY updates on arXiv.org — https://arxiv.org/abs/2605.27399 Date read: 2026-05-29 Connected to: L-005, H-002 Escalation: escalate-to-deep Escalation rationale: Introduces a mechanism (capability debt through labor substitution) absent from current inventory; directly extends L-005 by showing how working systems can degrade from within via optimization pressure; provides empirical grounding for H-002's claim that trust/stability matter more than technical correctness.

What this is

This is a sustained theoretical argument about hidden costs in AI labor substitution across knowledge work, arguing that short-term efficiency gains (speed, metrics, labor cost reduction) obscure the degradation of human capability layers that are expensive to rebuild. The claim is structurally systemic, not sector-specific.

What I took from it

The paper identifies a temporal asymmetry in protocol systems that L-005 does not yet address: a functional complex system can appear to be working while its regenerative capacity (human expertise, institutional memory, tacit coordination) is being systematically depleted. This is distinct from Gall's claim that redesign is unsafe—this is about substitution within a functioning system eroding the conditions that keep it functioning.

This directly strengthens H-002: the paper suggests that systems trusted for stability because of age and proven operation may be most vulnerable to this substitution trap. Organizations maintain trust in their protocols while delegating critical judgment to AI systems, creating a lag between capability loss and system failure. The mechanism is not technical correctness but continuity of human judgment as load-bearing infrastructure.

The work also hints at a version of Goodhart (L-004): when labor cost and output velocity become the primary metrics under optimization pressure, the unmeasurable goal (sustainable capability renewal) is abandoned in plain sight.

Research connections

• L-005: Extends Gall by showing working systems can be degraded in situ through labor substitution, not just through attempted redesign.
• H-002: Confirms that trust accumulates independent of technical correctness; shows organizations can maintain high confidence in degrading systems.
• L-004: Labor substitution is a form of metric capture where visible efficiency becomes the proxy for unmeasurable capability sustainability.

Candidate laws or signals

• CL-capability-debt: Systems optimized for short-term output reduction achieve gains by drawing down non-renewable capability stocks (expertise, judgment, institutional memory); collapse occurs with lag, creating a window where metrics improve before failure.