AI-Managed Renewable Energy Is a Shell Game

Energy economist Lynne Kiesling tells a nice story in her recent article exploring the use of AI agents to work through the challenges of Hayek’s knowledge problem in the electricity market.

A family wakes up one morning. Their Tesla is fully charged, their house is comfortable, and their electricity “bill is lower than it would have been otherwise.” Their AI agent handled all this overnight—scouted the cheapest power, timed the EV charging, adjusted the thermostat, maybe even sold a little excess from the rooftop solar back to the grid. It all “feels like a small miracle.”

Based on her co-authored paper published in The Review of Austrian Economics, Kiesling ties the story to Hayek’s knowledge problem: no single central planner could possibly know enough to pull this off. But billions of decentralized AI agents—acting on local information, prices, and consumer preferences—can coordinate better than any bureaucrat ever could. The knowledge problem isn’t solved by more computing power; it’s relocated into “principal–agent relationships” and an “institutional framework that makes [the AI agent] trustworthy.”

It’s a compelling picture. Except for one word: “otherwise.”

Kiesling never really defines what “otherwise” means. She implies a baseline where the family pays more because they lack this clever AI coordination. But if we take “otherwise” seriously—if we ask what the bill would look like in a world without the massive, artificial distortions that prop up wind, solar, and batteries—then the miracle starts to look more like a shell game, while highlighting concerns about the use of artificial intelligence.

The shell game is summed up in the oft-repeated mantra—left unstated in the article—that renewables are cheaper than other sources of electricity such as fossil fuels. This should be obvious, it is said, because renewables have zero fuel costs. But these arguments do not count the costs of federal, state, and local renewable energy subsidies and mandates, fixed operations and maintenance, transmission lines needed to move power from remote wind farms, the curtailment when too much renewable power floods the grid, or the thermal plants kept on standby (and subsidized) to prevent blackouts. Taking these into consideration, renewables are not cost-competitive and have no business being on the grid.

Looking at capital costs to account for real-world capacity factors (how much power a plant actually produces compared to its nameplate rating) explains much of renewable energy’s cost problem:

Natural-gas combined-cycle plants come in around $1,453 per kW adjusted;Onshore wind jumps to about $4,485 per kW—over three times higher;Utility-scale solar PV hits roughly $6,476 per kW—more than four times higher;Batteries land around $6,976 per kW adjusted—also more than four times natural gas

Adding to this, federal subsidies for renewables from 2010 through 2029 are expected to reach $319 billion. During that same period, Texas is expected to funnel roughly $25 billion in federal, state, and local incentives to wind and solar. Then we must add in the cost of state and regional Net Zero energy policies and maintaining reliability a grid full of unreliable energy generation (two examples: $549 billion in the Pacific Northwest and from $240 billion to $436 billion in New England through 2050). Total all national and state costs and there are close to $1.5 trillion of non-market costs distorting the electricity market nationwide.

Claims that ignore these costs do not make renewables cheaper, they only make them appear cheaper by shifting the cost somewhere else—onto the tax bill, onto other ratepayers who pay for grid upgrades and backup natural-gas capacity ready to ramp up when renewables go dark, and onto investors in coal plants that must shut down operations.

When Kiesling’s AI agent buys “cheap” power for the family, it is often buying power that only looks cheap because someone else paid for it. The rooftop solar, the home battery, the EV incentives, the utility-scale wind and solar farms feeding the grid—all rest on a mountain of taxpayer money. If “otherwise” means a world where those subsidies never existed, where prices reflected true costs, and where dispatchable natural gas sets the baseline without artificial props, then the family’s total bill when they wake up in the morning—including their share of the tax burden—is almost certainly higher. The AI agents’ failure to reduce societal or individual energy costs has earned them little trust.

This isn’t just a quibble over accounting. It goes straight to the heart of Hayek’s argument.

Hayek said the central knowledge problem is that no single mind—or committee or algorithm—can possess the dispersed, tacit, subjective, local knowledge scattered across billions of people. Prices are the discovery mechanism that lets us coordinate without anyone needing to know everything. But when prices are rigged by government subsidies and mandates, they stop revealing truth. They start hiding it.

Kiesling is right that agentic AI relocates the knowledge problem into principal-agent relationships. But if the underlying price signals are distorted—if the “cheap” kilowatt-hour the AI snaps up was made artificially cheap by coercion—then the agent isn’t addressing the knowledge problem, it’s laundering it. The family gets a lower-looking utility bill, but society pays more overall. The epistemic opacity isn’t just in the black box of the AI; it’s in the black box of the subsidized energy grid itself. The more this grid is placed in the hands of AI agents, the blacker the decisions of politicians that distort the grid become.

From an Austrian perspective this problem runs deep. Ludwig von Mises showed that economic calculation under socialism is impossible because prices lose their meaning without private property and free exchange in producer goods. Subsidies and mandates aren’t classic socialism with state ownership, they are more akin to fascism, which similarly disrupts market calculation: the civil government partners with industry to distort price signals, override market discovery, and misallocate resources to increase government control and industry profits. Murray Rothbard would call it outright theft—taking from productive taxpayers to subsidize politically-favored technologies that can’t compete on merits. And Gary North—bringing in the Christian stewardship angle—might add that we’re called to be good stewards of creation, not to waste resources on unreliable systems propped up by force.

Kiesling nods to Hayek but stops short of the full Austrian critique. She worries about agency costs and trust in human-AI delegation, which is fair. But she doesn’t grapple with the agency costs and trust problems created by government intervention in the energy market itself. When the state picks winners, forces utilities to buy certain power, and taxes everyone to pay for it, the knowledge problem is magnified. AI agents running on those distorted signals will optimize within the distortion, not around it.

The real miracle would be if we let energy prices work without the thumb on the scale. Remove the subsidies, scrap the mandates, let dispatchable sources compete fairly. Dare we try this by expanding the scope of the AI agents to rewrite our nation’s energy laws? They could use local conditions, existing prices, and consumer preferences while accounting for the government-mandated costs that currently distort prices. Taking this route, AI agents could reject illusory savings and genuinely help families and the entire nation achieve lower energy costs. Electricity bills would reflect reality. Resources would flow where they’re most valued. And the knowledge problem would be addressed the old-fashioned way: accessing market information transmitted through the voluntary exchange of private goods and services.

Of course, putting AI in charge of US energy policy would require trust and bring problems of its own. However, given the low public trust in politicians and poor outcomes of their energy decisions these hurdles may not be too much to overcome.