Does codifying expertise into AI agents drive faster labor substitution?
This explores whether turning tacit human expertise into the structure of an AI agent (its rules, scaffolding, memory) speeds up the replacement of skilled workers — and the corpus suggests the link is real but more conditional than a straight line.
This explores whether turning tacit human expertise into the structure of an AI agent — its rules, scaffolding, memory — speeds up the replacement of skilled workers. The corpus says yes, there's a mechanism, but it's gated by firm capability and coordination, not by the codification alone.
Start with the mechanism. The most direct evidence is an industrial case where domain rules and design principles were embedded into an agent's scaffolding, letting non-experts produce expert-rated output and dissolving the need for specialist oversight Can codified expertise let non-experts match specialist output?. The striking part is *where* the capability came from: not bigger models, but externalizing tacit knowledge into structured harness components. A companion finding generalizes this — reliable agents work by offloading memory, skills, and protocols into a harness layer rather than leaning on model scale Where does agent reliability actually come from?. So codified expertise is precisely the substance that makes an agent good enough to stand in for a specialist. That's the substitution engine.
But substitution doesn't happen at a uniform rate. Firms replace marketplace labor with AI at firm-specific speeds, and higher-exposed firms do it faster and cheaper — the returns accrue to internal AI capability, not to the technology being equally available to everyone Do firms substitute labor for AI at different rates?. Codifying expertise *is* that internal capability, which means the firms best positioned to build expert-laden agents are the ones that pull away. And the labor effect isn't only displacement: when AI exposure is concentrated in a few tasks, workers reallocate to non-displaced work, blunting aggregate job loss Does concentrated AI exposure enable workers to adapt and reallocate?. Whether codification drives *faster* substitution or just *reshuffles* tasks depends on how broadly the expertise spreads across a role.
There's a ceiling worth knowing about. Agents built from codified expert demonstrations inherit the limits of their curators — they can't learn beyond the scenarios someone imagined and encoded, because they never interact with live environments during training Can agents learn beyond what their training data shows?. So codification substitutes well for *routine* expert judgment but caps at the frozen edge of what was written down; the genuinely novel call still wants a human. And as these agents become economic actors, the binding constraint shifts from raw capability to coordination — whether agents can transact, settle accounts, and leave auditable trails When do agents need coordination more than raw capability?. That governance bottleneck can slow substitution even when the expertise is fully codified.
The part you didn't know you wanted: the deeper risk isn't speed, it's drift. One line of work argues societal systems stay aligned with human preferences partly *because* institutions depend on human workers who care about outcomes — and incrementally replacing that labor with agents erodes the implicit alignment, potentially irreversibly Does incremental AI replacement erode human influence over society?. So codifying expertise doesn't just substitute labor faster; it quietly removes the human dependency that kept the system answerable to humans in the first place.
Sources 7 notes
An industrial case study embedding domain rules and design principles into an LLM agent's scaffolding achieved 206% output-quality improvement and expert-level ratings from non-experts, bypassing the need for specialist oversight. The capability gain came from externalizing tacit expertise into structured harness components, not from model scale.
Research shows reliable LLM agents externalize three cognitive burdens—memory (state persistence), skills (procedural components), and protocols (structured interaction)—into a harness layer rather than relying on model scale alone. The harness unifies these externalities and eliminates the need for the model to solve the same problems repeatedly.
Higher AI-exposed firms replace online labor marketplace workers with AI tools faster and at lower cost than less-exposed firms, suggesting returns to scale in internal AI capability rather than uniform technology diffusion.
Analysis of task-level AI exposure across firms 2010-2023 shows that while higher mean exposure reduces labor demand, more concentrated exposure (affecting few tasks) enables workers to reallocate to non-displaced tasks, producing modest net employment effects.
Agents trained on static expert datasets cannot learn from their own failures or generalize beyond demonstrated scenarios because they never interact with environments during training. Competence is capped by what curators imagined, not by agent capacity.
Once agents hold credentials, transact value, and interact with other agents, raw model capability stops being the limiting factor. The real bottleneck becomes whether agents can coordinate reliably, settle accounts, and leave auditable evidence of their actions.
Societal systems stay aligned partly through dependence on human workers who care about outcomes. As AI replaces this labor, explicit alignment controls weaken and systems drift from human preferences. Interdependent misalignment across institutions could become irreversible.