As AI drives the cost of generating competent work toward zero, organizations gain more options but not automatically more value. Research shows productivity gains in specific tasks alongside convergence in AI-assisted output and reduced accuracy outside the model's capability range. The strategic advantage is shifting from production capacity to judgment: choosing what deserves to exist and taking responsibility for the decision.
There used to be a useful constraint in creative and technical work: producing something took time. A new campaign required writers, designers and production. Even a mediocre idea had to survive a budget and a calendar before it reached the world.
AI is weakening that constraint quickly. Stanford's 2025 AI Index reported that the cost of querying a model performing at roughly GPT-3.5 level fell more than 280-fold between November 2022 and October 2024. The price of competent output is moving toward zero. Stanford AI Index
That sounds like uncomplicated progress. In many ways, it is. But when output becomes cheap, companies do not automatically create more value. They create more options. The difficult work moves from producing an answer to deciding which answer deserves to become real.
The filter is moving
An NBER study of more than 5,000 customer-support agents found that access to a generative AI assistant increased issues resolved per hour by 14 percent on average. In a six-month experiment across 66 companies, workers who adopted the AI tool spent two fewer hours on email each week. Generative AI at Work · Shifting Work Patterns with Generative AI
These are useful improvements. They are mostly improvements in execution.
A company can now produce ten campaign concepts instead of three and prototype software at a speed that would have looked unrealistic a few years ago. However, none of that answers the harder questions. Which of those concepts reflects what the company believes? Which one solves a problem people actually have?
The old filter was production capacity. The new filter is judgment.
Good enough is becoming abundant
This shift has an uncomfortable side effect. AI is good at producing work that looks competent. It can create a clean presentation, a plausible strategy and polished copy in seconds. I think this makes poor ideas harder to spot, because they no longer arrive looking unfinished.
There is evidence of this convergence. In a peer-reviewed study of creative writing, access to generative AI ideas improved the quality of individual stories, especially for less creative writers. But the AI-assisted stories also became more similar to one another. Science Advances
Fluent output can be wrong. When 758 management consultants used AI in a Harvard Business School experiment, they worked faster on problems inside the model's capability range but were 19 percentage points less likely to answer correctly on a problem outside it. I think the output looked just as polished either way. HBS Working Paper
That is not an argument against using AI. It is a warning about accepting its first reasonable answer.
When everyone has access to similar models, competence becomes easier to reproduce. A polished result no longer tells us much about the thinking behind it. The advantage moves to whoever can see that an answer is technically correct but strategically wrong or unnecessary.
In Pixar's *WALL-E* (2008), the Axiom has solved production entirely. Food, entertainment and comfort arrive without effort. Humans float through their days on hover chairs, consuming whatever the ship generates, having long since stopped choosing anything for themselves.
The problem aboard the Axiom is not scarcity. It is the absence of judgment. Nobody decides what matters, so nothing does. When a single living plant arrives on the ship, the autopilot tries to suppress it, not because the plant threatens anything but because AUTO is enforcing Directive A113: a standing order never to return to Earth. The system cannot reconsider an outdated rule. The captain has to relearn what it means to override a directive and make a choice that matters.
That is the version of AI adoption worth worrying about: not the machines producing too little, but everyone accepting too much.
Taste is practical judgment
Taste can sound like an artistic luxury. I think it will become much more practical than that.
Taste is the ability to frame the right problem before generating solutions. It means understanding enough about people, technology and context to recognize why one reasonable answer is better than another, and knowing when to throw away work that took only seconds to produce.
This is where accountability enters. An AI system can propose an action, but someone still needs to decide whether it serves the customer, the company and the intended outcome. Generating the option is becoming easy. Taking responsibility for the choice is not.
Companies will need to design for this deliberately: clearer principles, stronger review practices and people with enough domain knowledge to challenge plausible output. Otherwise, lower production costs will simply create more noise for the organization to process without value gained.
The companies that learn to choose
The first phase of generative AI focused on speed: faster writing, faster coding, faster analysis. The next phase will expose a more important difference between companies.
Some will use AI to multiply everything they already produce. Others will use it to become more selective. They will explore more possibilities while becoming stricter about what reaches customers, enters production or shapes a decision.
I believe the second group will build the stronger organizations.
AI gives us an extraordinary expansion of productive capacity. But capacity alone has never been the same as direction. As creation becomes cheaper, the ability to choose with clarity, conviction and responsibility becomes more valuable.
The defining capability of the AI era may not be how much we can generate. It may be how well we decide what should exist.
