Anish Acharya and David Pawlan show why AI agents must outgrow the twenty-dollar daily subsidy
By Bram de Vries · Reporting from Amsterdam ·
Podcast
It costs twenty dollars a day to give one person a truly ambitious AI agent. Anish Acharya of a16z noted this figure on the latest "a16z" podcast. This is not a business model. It is a venture capital subsidy for the impatient. David Pawlan joined him to discuss the explosion of 122 different AI assistants. The market is currently a wild west of proactive software. These tools do not just wait for orders. They book flights and file for tax refunds. They are moving from being assistants to being agents with actual agency.
The venture capital burn for invisible labor
Acharya estimates the cost of running these agents at hundreds of millions of dollars annually for a single startup. This is the price of seeing God in the code. Pawlan has tested dozens of these tools through his site, Assistant Benchmark. He found that 65 of the 122 assistants he tracked are already charging money. The market is trying to find what a human will actually pay for. Most people do not care about being ten percent more efficient. They care about the invisible employee who handles the drudgery. Pawlan described biking to work while his agent cleared his inbox to zero. This is the promise of the technology. It grows the economic pie by reclaiming wasted hours. But the current cost of tokens makes this a game of chicken. Startups are betting that compute prices will crash before their bank accounts do.
Amazon defends its ad-revenue toll booth
The most revealing conflict is between the platforms. Amazon recently blocked the Muse agent. Shopify embraced it. Acharya and Pawlan noted that Amazon’s business model relies on ad revenue. If an agent buys your socks, you do not see the ads. Amazon is acting as the incumbent protecting a legacy profit pool. They are prioritizing their toll booth over consumer efficiency. Shopify wins by moving more goods. They do not care if a human or a yeti-shaped bot clicks buy. This is the mercantile tradition at its best. The friction of the commerce web is a tax on the consumer. Agents threaten to abolish that tax. The strongest case against this is that consumers enjoy the hunt. Critics like Ben Thompson argue people want to spend time, not save it. They say shopping is entertainment. This ignores the millions of people who find administrative life a burden. The historical pattern is clear. Efficiency always wins over entertainment that is actually just friction.
The individual versus the bureaucratic machine
Citizens should look at these agents as a way to bypass bureaucracy. Acharya pointed out that 1.5 million people get DUIs every year in the United States. Most lack access to the family office style of legal and admin support. AI agents could democratize that high-end service. They could handle the bureaucratic pressure that weighs on the average household. This is not just about productivity. It is about power. An agent that acts on your behalf is a tool of individual liberty. It removes the need for the state or the giant corporation to hold your hand. The risk is trust. Pawlan warned that if an agent crosses the line once, the user is gone. Trust is the only currency that matters in trade. If an agent breaks up with your girlfriend or hallucinates your name on a flight, the contract is void.
The era of the free-spending AI experiment is ending. We are moving toward a consumer-aligned internet where we pay for results. The winners will not be the ones with the best personality. They will be the ones that save the most money and move the most cargo. The market has no room for twenty-dollar-a-day hobbies. It demands tools that earn their keep.