Perspective_

Much of our careers have been spent thinking about software as a way to make people more productive.

We build applications, dashboards, workflows, and processes that take something complicated and turn it into something people can use. It works remarkably well. We encode expertise into software, establish a process, and allow people to execute it.

But there is an interesting tradeoff we don't talk about often. The more expertise we encode into our systems, the less thinking we require from the person using them.

A dashboard tells us what to look at. A workflow tells us what to do next. An application gives us the fields to complete and the choices to make. In many cases, the difficult thinking has happened before the user ever interacts with the system. Someone decided what information mattered, which decisions were likely to be required, and what sequence of actions should follow.

That is not a flaw in traditional software. For decades, one of technology's greatest achievements has been its ability its ability to take expertise and embed it into systems that allow more people to execute it. Software reduces friction by reducing the number of questions we have to ask at the point of execution.  

Conversational AI changes that relationship.

When we work with an AI system, there often isn't a predefined workflow waiting for us. We have to explain what we’re trying to accomplish, provide context, establish constraints, challenge its assumptions, evaluate its response, and refine the question.

The conversation itself becomes part of the work. In that sense, AI puts something back into the interaction that decades of software design worked to remove: the need to think through the problem.

That has caused us to question one of the prevailing narratives about AI—that increasingly capable systems will make us think less. They certainly can. If we treat AI as an answer machine, accept its first response, and move on, we are outsourcing judgment. AI can make intellectual passivity remarkably efficient.

But used differently, AI can do almost the opposite. It can make thinking cheaper.

Historically, the bottleneck wasn't necessarily knowing what we wanted to explore. It was the cost of exploring it. An idea had to become requirements, then designs, then code, then tests. It had to move between people and teams. Even when doing the work ourselves, a tremendous amount of cognitive energy went into translating an idea into something executable.

Execution imposed discipline. Building something required time, money, expertise, and coordination. We couldn't pursue every idea, test every hypothesis, or explore every possible solution because the cost of doing so was too high. The difficulty of execution forced choices upstream.

AI changes that equation. We can explore an architecture, build a prototype, test an idea, analyze a dataset, or challenge an assumption in a fraction of the time. We’re not necessarily thinking less. We’re spending less time translating thinking into artifacts and more time thinking about the problem itself.

The most significant impact of AI may not be that it allows us to execute known tasks faster. It may be that it lowers the cost of exploring things we don't yet know how to do.

When the distance between an idea and a prototype collapses, we can explore more possibilities before committing to one. We can challenge assumptions instead of accepting them. We can move more fluidly between thinking and building, learn from what we create, and bring that learning immediately back into the next question.

Lowering the cost of execution does not lower the value of judgment. It increases it.

AI can generate ten potential approaches instead of one. Someone still has to decide which approach is worth pursuing. It can rapidly build a prototype. Someone still has to determine what problem the prototype should solve. It can analyze more information than any individual could reasonably process. Someone still has to recognize which question matters in the first place.

For decades, organizations have been constrained by what they could afford to build. Increasingly, they may be constrained by something more fundamental: what is worth building at all.

That may be one of the most important shifts AI creates. As the technology removes friction from execution, humans become increasingly responsible for providing the things the technology cannot determine for itself: which problems matter, which tradeoffs are acceptable, what constraints should apply, and when an answer is good enough to act on.

Those questions become even more consequential as AI moves beyond helping us explore ideas and begins participating more directly in decisions and actions.

The more capable these systems become, the more we may depend on them to participate in our thinking. And the more we depend on them, the more important it becomes to understand what we are thinking with and where the boundaries of our trust should be.

When an AI recommends a decision, how do we know what to validate? When it generates a solution, how do we evaluate it? When it is allowed to take action, what constraints should govern that action?

The answer is not necessarily to force AI back into the deterministic model of traditional software. Humans aren't deterministic either. We have learned how to build organizations and systems that account for human uncertainty, establish appropriate controls, and create mechanisms for accountability.

AI changes that challenge, but it does not eliminate it.

As AI becomes more capable and more embedded in how we explore and act, we will need better ways to evaluate it, monitor it, validate its outputs, and establish clear boundaries around what it is allowed to do. Not because uncertainty can be eliminated, but because increased capability does not remove the need for judgment and responsibility.

And perhaps that is where the conversation about AI's impact on work has become too narrow.

We often ask what AI will allow people to do faster, or which tasks it will eliminate altogether. But the more interesting question may be what people can do with the capacity created when the distance between thinking and doing begins to collapse.

We could use that capacity to complete the same work with fewer people. Or we could use it to investigate more problems, challenge more assumptions, test more ideas, and give more people the ability to turn an insight into something real.

The better AI becomes at answering our questions, the more important our questions become.

Perhaps the real opportunity isn't to use AI so that humans have less to do. It is to use AI to remove more of the gap between thinking and doing—so that people can spend more of their capacity deciding what is actually worth our time.