By Dennis H. Goodhart, with Hymie, my AI collaborator
By now, I’m sure we have all heard about the hundreds of billions of dollars AI companies are prepared to spend building more powerful systems.[1]
Simple economics says that eventually those investments have to produce a return. That can come from attracting more paying customers, selling higher-priced services, increasing business use, or finding less expensive ways to deliver AI. From the customer’s side, however, the question is much simpler: What am I getting for the additional money?
- A smarter AI can be worth more if it saves time, improves the quality of my work, or lets me do something I couldn’t reasonably do before.
- But the real cost includes the time I spend checking, correcting and repeating its work.
- If AI companies ask customers to pay more, they will have to show what the additional money actually buys.
I Want It Done Right the First Time
I have an AI assistant that can help me research complicated real estate questions, examine competing arguments and write about artificial intelligence. Occasionally, it also needs reminding that we already tried something and it didn’t work. If you’ve ever spent an afternoon following an AI assistant through troubleshooting steps only to find yourself back where you started, you may recognize the experience. At that point, I’m not particularly interested in whether the next generation of AI can give me the same instructions faster. I want it done right the first time. Or, when that isn’t possible, I at least want it to remember what happened an hour ago and use that information to make the next attempt better.
AI can save me hours, days or even weeks on one assignment. On another, it can send me down a rabbit hole that consumes much of the time I thought I had saved. That is why I don’t measure productivity by how quickly AI produces an answer. I count the time until I have something I can actually use. If AI produces a report in five minutes and I spend another hour tracking down unsupported claims, correcting incorrect information or reworking the output because it won’t function with another system I use, the five-minute turnaround tells me very little.
What Would I Pay More For
As a real estate broker, I would pay more for a tool that could pull together the relevant zoning provisions for a property, show me exactly where the information came from and identify what still needs professional confirmation. The same goes for comparable sales. I want enough information to determine whether the properties are actually comparable—not simply because they are nearby or have similar square footage.
The system should also distinguish clearly between what current market data show, what assumptions are being made, what represents a forecast or projection, and what still needs to be independently verified before I advise a client. That could catch something important, save me hours of research and perhaps help move a transaction forward more quickly. That is added value I can understand.
I would also pay for an assistant that gets better at working with me. If we try a particular fix and establish that it doesn’t work, I want the AI to carry that lesson into the next attempt instead of eventually leading me back to the same solution. Researchers are already exploring ways for AI agents to reuse strategies learned from previous successes and failures. Google’s ReasoningBank research is one example.[2] My test is much simpler: Does the next job take fewer tries?
The Customer Gets a Vote
After more than four decades working in technology, I have seen how easily a company’s ideas about growth can get ahead of what its customers actually need, will use and will pay for. There is an old expression in technology: If it ain’t broke, don’t fix it.
That doesn’t mean we should stop improving technology. It means an improvement has to solve a problem somebody actually cares about. Some advances are obviously transformative. Moving from horse-drawn transportation to automobiles changed what people could do. Automatic transmissions made driving easier for millions of people. But not every improvement is transformative.
Is producing a report in ten minutes instead of twenty minutes enough reason for me to pay substantially more? Probably not. Producing that same report in ten minutes and getting it right, so I don’t spend the next two hours reviewing it, correcting errors and deciding whether I can trust it? That is something I would consider paying for.
If an AI company asks me to pay more, the amount it spent building the system gives me very little reason to say yes. I am not buying their data center. I am buying a service.
Some customers will gladly pay more for capabilities they use every day. Others may look at a higher monthly bill, look at a list of impressive new features they rarely use, and ask: Who told them to spend all that money on services we don’t need?
I already receive real value from AI, and I am open to paying more for improvements I can see in my workday. Help me complete a difficult assignment with greater confidence. Give me better information. Help me distinguish fact from projection. Remember what we already tried. Give me back time I can spend working with clients instead of correcting the tool that was supposed to save me time.
Do those things, and I can decide what the improvement is worth. That, ultimately, is the marketplace test: What does it do for the person paying the bill?
Sources
[1] J.P. Morgan Global Research, “What Unprecedented AI Capex Means for Markets and Investors,” October 19, 2026. J.P. Morgan projects approximately $900 billion in global AI capital expenditures in 2026.
[2] Ouyang et al., “ReasoningBank: Scaling Agent Self-Evolving with Reasoning Memory,” September 29, 2025. Research on enabling AI agents to reuse strategies derived from previous successful and failed experiences.
AI for the Perplexed is a collaboration between Dennis H. Goodhart and Hymie, my AI writing collaborator. Dennis retains editorial responsibility for the published content.
Disclaimer: The opinions expressed are those of the author. The author is not employed by, compensated by, sponsored by, or affiliated with any of the companies or products mentioned in this article. The author does have a paid ChatGPT subscription. ChatGPT, referred to here as “Hymie,” is credited as a writing collaborator.