Automation and Rent Dissipation: Implications for Wages, Inequality, and Productivity
Source: https://www.semanticscholar.org/paper/b0fc97b50af789df40bf92cc23a257d3168307ca ↗
Full text: open-access via OpenAlex ↗
Acemoglu and Restrepo solve a puzzle that haunts every conversation about AI and work: if automation increases productivity, why don't wages rise accordingly? Their answer is rent dissipation — automation systematically targets the highest-paid tasks within job categories, destroying the economic rents that allowed some workers to earn above their market value.
The result is a cruel efficiency: machines get more productive while humans get more precarious, not because technology is inherently anti-worker, but because it targets exactly the inefficiencies that workers had captured as higher wages.
For product directors, this reframes the automation question from 'what can we automate?' to 'what rents are we about to destroy, and do we want to?' The paper's empirical rigor — explaining 52% of rising inequality since 1980 — makes it essential reading for anyone building AI products who wants to think seriously about their distributional consequences.
Central argument
Acemoglu and Restrepo argue that automation disproportionately targets high-rent tasks — jobs where workers earn wages above their opportunity cost due to bargaining power, unions, or regulations — thereby dissipating those rents and amplifying wage losses beyond what standard displacement effects would predict. Using US data from 1980 to 2016, they find that this rent dissipation mechanism accounts for roughly one-fifth of the relative wage decline for automation-exposed groups and explains 10 of the 52 percentage points that automation contributes to rising between-group inequality. Most strikingly, because firms automate tasks where labor is artificially costly rather than where automation is socially most valuable, the inefficiency offsets 60–90% of automation's productivity gains, leaving net TFP growth from automation at only 0.3–1.3% over 36 years.
Critique
The model treats worker rents as exogenous wedges — a deliberate simplification borrowed from the misallocation literature — but this sidesteps the question of whether those rents themselves reflect productive functions: information asymmetries, retention incentives, or implicit contracts that sustain firm-specific investment. If high-rent jobs were partly sustaining tacit knowledge accumulation or organizational stability, then 'rent dissipation' may proxy for a deeper destruction of organizational capital that the productivity accounting does not fully capture. The paper's quantitative conclusions about welfare loss therefore depend heavily on a framing where rents are pure distortions, which is contestable.
Why it matters for product
For a CPO, the paper's core mechanism reframes a common automation rationale: teams often justify automating workflows by targeting the most expensive or friction-heavy human tasks, which maps precisely onto the 'high-rent task' pattern the authors identify as inefficient. If the costliest tasks in your product organization — senior judgment calls, cross-functional negotiation, nuanced discovery work — are expensive partly because they carry genuine informational or coordination value, automating them first may dissipate organizational capability while delivering far less productivity gain than projected. This also has implications for how AI-augmentation bets are prioritized in roadmaps: optimizing for cost reduction in high-wage activities is not the same as optimizing for value creation, and conflating the two can degrade the product team's strategic capacity over time.