AI 的迈达斯困境：当代码免费，功能便一文不值

Every team that adopted Claude Code, Cursor, or Copilot to produce more code has the same problem: When output is free, offering volume stops being a viable strategy. In fact, teams accelerating their software development with AI may be giving up the very thing that makes their software valuable.

It’s time to remember what makes software valuable and why racing to produce a mountain of code and features leads to failure.

The Midas touch

In Ovid’s tale, Midas, king of Phrygia, took in the missing satyr Silenus, whom Phrygian peasants had found drunk, granted him ten days of hospitality, and then returned him to Dionysus. In return, he was offered the reward of his choice: whatever he wished for. Famously, he wished that everything he touched would turn to gold.

Testing his new ability on an oak twig and then a stone, Midas was delighted. Nathaniel Hawthorne, who retold the myth for children in 1851, added a rose garden: His Midas turns every bloom to gold, still delighted, though his daughter Marygold mourns that the roses no longer smell sweet. In both versions, the celebration ends at the table, where food and drink turn to gold in his hands. That’s when Midas realized his reward was a bane, not a blessing, and asked to be rid of it.

Many leaders who force-fed AI to developers are now coming to a realization similar to that of Midas.

Though the process is gradual, many organizations and leaders who force-fed AI to developers are now reaching a similar conclusion. Reports show organizations slashing budgets and limiting tool use.

With Microsoft and Meta reportedly cutting internal spending on Anthropic’s Claude, and Uber and Accenture capping what employees can spend on AI tools, it’s time to acknowledge that the return on investment might not be there — McKinsey’s own 2026 research found record AI spending hasn’t moved earnings for 94% of enterprises — and it’s easy to see why.

The economics of plenty

If Midas had depended on the value of gold, perhaps hoping to trade the gold he created to fund his lavish lifestyle of hosting ten-day-long parties, he would have quickly learned that his gold was no longer valuable. His touch would cause hyperinflation, and gold’s price would plummet. His skill would be as useful as turning everything you touch to ash.

In the modern, non-mythical world, everything you touch with AI becomes a commodity. If you can get Claude or ChatGPT to create something for you, then everyone else can. They have access to the same tools. Whether you’re building software or writing a book, when you use AI, its perceived value is little more than the cost of the tokens you used.

When coding looked like magic to most people, basic coding competence was sufficient value. You could build a career translating human-language requirements into code. This value rolled up to the team, department, and organization level, with code magicians performing their arcane translations in return for money, whether the software was internal, built for a customer, or sold as a product.

If your software expresses user requirements translated into code, it’s now worthless.

If your software expresses user requirements translated into code, it’s now worthless. We know that a page that adds nothing beyond the AI summary is worthless, and software built by translating a user’s wants into code is the same.

How to avoid incinerating your software’s value

There is hope for us, though. When Midas asked to be rid of his power, he had to wash in the Pactolus River. Our touch isn’t directly connected to the generation machine, so it won’t inevitably turn our work to ash.

The answer is conceptually simple, though it takes a lot of effort in practice, which seems to be exactly what people want to avoid. Nevertheless, for your work to be immune to AI’s mortally ruinous touch, you need to provide something beyond what someone could achieve with their own Claude subscription.

For your work to be immune to AI’s mortally ruinous touch, you need to provide something beyond what someone could achieve with their own Claude subscription.

That means moving away from the feature-factory model of software development, where the value is primarily in the translation of “human wants” into code. Instead, teams building software must develop a deeper understanding of the jobs users do and the challenges they face, so they can generate novel ideas and commercial insights users wouldn’t reach on their own.

This second mode of building means you may deliver something completely different from what users ask for, yet one that gracefully solves many of the challenges they face in their work. Combining deep business domain knowledge with an understanding of abstract, reusable solutions lets you deliver something exceptional.

DORA’s State of DevOps research has highlighted this need for user-centricity. That doesn’t mean doing whatever users ask; it means understanding their work and challenges so you can provide software they value. DORA’s 2024 report says this approach improves product performance, regardless of delivery speed.

We are living through a time of excess code, sentences, and images. The path forward isn’t more; it’s better.

The post AI’s Midas problem: When code is free, features are worthless appeared first on The New Stack.