The Agentic Jevons Paradox: Why Cheap AI Paperwork is a Trap

An organization can save money drafting a document while making everyone who receives it more expensive to operate.

Autonomous agents could turn that familiar irritation into an institutional crisis. Legal objections, compliance questionnaires, regulatory comments, policy briefs, and internal risk reports are all becoming drastically easier to produce. But every time a generative system drafts one of these documents, a human recipient acquires another item to classify, verify, answer, or defend against.

The savings appear in the sender’s budget. Much of the work appears in someone else’s.

The result could be an extraordinary expansion of bureaucracy, ironically driven by the very organizations that bought AI to reduce it. I call this the Agentic Jevons Paradox.

The Economics of Cheap Paperwork

In 1865, economist William Stanley Jevons observed that improvements in the efficiency of coal use didn’t reduce coal consumption — they expanded its applications and increased total demand. Efficiency changed what people could afford to do.

Institutional text has a similar demand mechanism. Drafting a lengthy filing or assembling a regulatory campaign once required substantial skilled labor. Lower drafting costs make additional filings, objections, reviews, and requests worth attempting.

The feedback loop is straightforward: An organization automates documentation and produces more of it. Its counterparties automate their responses. Each side discovers further questions to ask, exceptions to investigate, and risks to document. Some of this activity is useful. Much of it exists simply because producing it has become cheaper than deciding whether it is actually necessary.

The immediate problem is the cost imposed on the recipient. A sender may spend fractions of a cent generating a request that takes a specialist an hour to resolve. If the sender receives the benefit while the recipient bears the review cost, individually sensible automation produces collectively excessive paperwork.

The danger is broader than a rebound in software spending. Cheap generation transfers costs into scarce review capacity.

The Threat of Regulatory Flooding

Imagine a competitor generating thousands of objections about your company’s supply chain, tailored to different products and jurisdictions. Some are duplicates in different languages. Some are weak but require legal checking. A few identify real problems.

You might use agents to sort the material and draft replies. Regulators might do the same. Yet disputed facts, significant commitments, and consequential decisions still require accountable human review. The bottleneck moves from producing a response to deciding which responses can safely be approved.

We already have precedent for volume manipulation. In its 2017 net neutrality proceeding, the FCC received over 22 million comments. A subsequent investigation concluded that nearly 18 million were fake. While public comments differ from strict court filings, the episode proves that open institutional processes attract manufactured participation at an enormous scale. Generative systems will only make such submissions more varied, localized, and much harder to filter via exact-match duplication.

Defending the Review Queue

How do institutions survive an environment where the cost of generating a plausible claim drops to near zero, but the cost of verifying it remains high?

A filter that successfully removes 90 percent of incoming material is no longer sufficient if inbound volume increases by a factor of a hundred. Organizations will need to adopt what I call Sovereign MLOps — retaining strict institutional control over sensitive data, inference, memory, and consequential actions — while deploying aggressive, semantic claim-compression architectures to group incoming disputes without erasing valid dissent.

Read the full analysis on Vector1 Research

If you want to dive into the mechanics of how we actually build these defenses, I’ve published the complete breakdown over on my Substack.

In the full article, I explore:

• The Information Theory of overload: A mathematical framework for measuring your organization’s true review burden and saturation ratio.
• The consequences for institutional scale: Why this dynamic favors large incumbents and forces risky delegation to automated agents.
• A Python/spaCy Prototype: Open-source code for a conservative claim-grouping architecture that uses Named Entity Recognition (NER) and Natural Language Inference (NLI) to compress repetitive claims without falsely merging critical disagreements.

Read the full article and technical appendix on Vector1 Research here https://brianjamescurry.substack.com/p/the-agentic-jevons-paradox

About the Author

Brian Curry is a Kansas City-based machine learning and AI strategist, marketing economist, and applied researcher. He leads Vector1, an independent research studio working on cognitive AI, causal inference, marketing economics, and knowledge engineering, with a focus on simulation-driven intelligence systems.

Over more than twenty years, Brian has led data science and analytics initiatives at enterprises ranging from $4B to $125B in revenue, including Koch Industries, Tractor Supply, Vail Resorts, Hallmark, Garmin, and AT&T. His work applies economics and causal inference to connect marketing investment to structural outcomes, operational decisions to economic value, and AI adoption to durable competitive advantage.

He is also the founder of KC AI Lab, a 1,500-member AI and ML education community in the Midwest. He lives in Kansas City and is an active part of the Kansas City community.

Research and Open-Source Projects

The modeling approach in this paper draws on PyCausalSim and Papilon.

Causal Inference and Economics

• PyCausalSim: Simulation-driven causal discovery and evaluation, using counterfactual scenarios to isolate structural drivers.
• Papilon: Complex-system simulation and optimization, with applications in media mix modeling and adaptive architectures.
• Daedalus: Causal analytics for money-flow mapping and macroeconomic tracking.
• The Autonomy Economy: A macroeconomic framework for the combined effects of rapid productivity growth and structural price deflation in an AI-automated economy.

Forecasting and Applied Modeling

• Regimecast: Time-series regime detection, structural breaks, and economic forecasting.
• CarryNet: Financial modeling for commodity carry trade analysis.
• Foodmesh-ml: Machine learning, optimization, and simulation for distributed food systems.

Cognitive AI and Semantic Architecture

• Memory-Node Encapsulation (MNE): A cognitive AI architecture and data structure for artificial episodic memory (NeoCortex-M).
• MeaningFlow: Semantic content modeling, embeddings, clustering, and coverage gap analysis.

Contact: brian@vector1.ai | GitHub @Bodhi8 | Substack

The text of this article may be shared with attribution under Creative Commons Attribution 4.0 International (CC BY 4.0). Code and models referenced here are available separately to paid subscribers.