The Problem with Getting the Answer Too Early

I’ve noticed something changing in the way we approach problems at work.

A few years ago, if I wanted to understand why a particular business wasn’t scaling, I would spend considerable time working through the possibilities.

Was it a positioning problem?

Was customer acquisition becoming too expensive?

Was the product attracting the wrong customers?

Or was the company simply trying to grow faster than its operations could support?

I’d read, compare businesses, revisit assumptions, and occasionally spend hours pursuing an explanation that turned out to be wrong.

Today, I can describe the same problem to an AI model and receive a reasonably sophisticated initial diagnosis in seconds. That’s an extraordinary improvement, but I’ve been wondering whether something important happens during those hours we no longer need to spend.

Sometimes, the most useful insight comes from something you discover while trying to find the answer.

You begin to recognize patterns, understand relationships between seemingly unrelated problems, and develop a sense of which assumptions deserve closer examination.

There’s a kind of friction involved in this process that I think is worth preserving. I call it Productive Friction, the effort that develops understanding, judgment, or capabilities that remain useful long after the immediate task is complete.

We’ve spent decades trying to eliminate friction from work, and for good reason. But as AI makes it possible to bypass increasingly complex intellectual tasks, I think we need to become more selective about which difficulties we eliminate.

Some of them may be doing more for us than we realize.

Not All Friction is Waste

A finance team shouldn’t spend hours consolidating spreadsheets. A sales executive shouldn’t have to manually update twenty CRM fields. Removing these inefficiencies gives people more time for valuable work.

Now consider a strategy associate investigating why a company’s revenue is growing while profitability is declining.

They examine customer segments, margins, acquisition costs, and discounting. After working through several possibilities, they discover that the fastest-growing customer segment is also the least profitable.

The immediate result is a diagnosis. But the associate has also developed an understanding of how growth can conceal deteriorating economics, something they may recognize much faster the next time they encounter a similar problem.

If AI provides the diagnosis immediately, the company saves time. Whether the associate develops the same understanding depends on how they engage with the analysis.

This is why I find it useful to distinguish between two forms of friction:

Operational friction: Effort that consumes time without meaningfully improving outcomes or developing useful capabilities.

Productive friction: Effort that builds understanding, judgment, or skills valuable beyond the task at hand.

Both can exist within the same workflow. The challenge is figuring out which activities are worth preserving as we automate more of our work.

The Difference Between Producing Work and Developing Capability

In a 2025 study, researchers from Microsoft and Carnegie Mellon surveyed 319 knowledge workers about 936 examples of AI-assisted work.

They found that workers who reported greater confidence in AI’s ability to perform a task also tended to report less critical-thinking effort. Workers who were more confident in their own abilities reported greater engagement in critical thinking.

The researchers also observed a change in the nature of intellectual work. People increasingly spent their time verifying information, integrating AI responses, and supervising outputs.

The study relied on self-reported experiences, so it doesn’t establish that AI causes a decline in critical-thinking ability. Still, it raises an interesting question about how people will develop judgment when more of the underlying analysis is performed for them.

Another paper I found particularly interesting is Navigating the Jagged Technological Frontier, which examined 758 Boston Consulting Group consultants.

On tasks within GPT-4’s capabilities, consultants using AI worked 25.1% faster and produced higher-quality results. But when presented with a complex problem outside the model’s capabilities, AI-assisted consultants were 19% less likely to arrive at the correct solution.

The experiment highlights the importance of recognizing when an apparently convincing recommendation may be unreliable.

Consider two managers preparing market-entry strategies. Both use AI and produce impressive reports. One has spent years studying markets, testing assumptions, and making difficult commercial decisions. The other has much less experience and relies heavily on generated recommendations.

Their reports may look remarkably similar. Yet their ability to recognize a flawed assumption could be very different.

AI can narrow the visible gap between experienced and inexperienced professionals without necessarily closing the underlying gap in judgment.

For businesses, that creates a problem worth paying attention to. We can now improve the quality and speed of employees’ output without necessarily improving their ability to independently evaluate it.

The Hidden Return on Difficult Work

When we evaluate productivity, we usually look at immediate results: time taken to complete the task, the cost involved, and the quality of output.

But some intellectual work produces a second return that rarely enters the calculation.

For instance, consider a founder researching a new market.

AI can assemble competitor analyses, pricing comparisons, customer segments, and distribution recommendations in a fraction of the time previously required. Yet the founder might learn considerably more by speaking to potential customers, examining competing businesses, and developing an initial hypothesis.

They may discover that customers describe the problem differently from the industry. Perhaps the most successful competitor is targeting an unexpected segment, or the original market-entry question turns out to be wrong.

These discoveries shape how the founder understands the market, and that understanding can influence decisions for years.

I think strategically important work often produces two returns:

The Immediate Return: The value of completing the task.

The Capability Return: The value of becoming better equipped to handle future problems.

The second is difficult to quantify, which makes it easy to overlook when calculating the benefits of automation.

A task that takes four hours may seem expensive when measured against an immediate output. But if the experience improves twenty future decisions, the economics look rather different.

This is particularly relevant to executives developing their next generation of managers. Much of the judgment we expect from experienced leaders comes from years of investigating problems, making decisions, and learning from outcomes.

As those activities become easier to automate, organizations will need to be more deliberate about how that experience is developed.

Making AI Part of the Learning Process

There is encouraging evidence that AI can also accelerate learning.

In a 2025 Harvard study, students using a specially designed AI tutor learned significantly more in less time than students receiving conventional active-learning instruction.

The tutor encouraged participation, provided targeted feedback, and adapted to individual learning needs.

I find this particularly relevant because it shows how much the design of the interaction matters.

Take a manager working through a difficult pricing decision.

They could ask AI to analyze the market, evaluate competitors, and recommend a pricing strategy. That may be an efficient way to reach an initial answer.

Alternatively, they could begin by developing their own hypothesis, then use AI to challenge their assumptions, identify missing evidence, and explore competing explanations.

The second approach creates opportunities for the manager to exercise and refine their judgment while still benefiting from AI’s analytical capabilities.

For recurring, strategically important problems, this is an approach I would encourage. The additional effort needn’t be substantial. Sometimes spending fifteen minutes forming an independent view before reviewing AI’s recommendations may be enough to make the exercise more valuable.

The Productive Friction Test

In my organization, there’s an exercise we often undertake before automating an intellectual task. We ask three questions.

1. What capability does this task develop?

Be specific. Instead of identifying something broad like critical thinking, consider capabilities such as diagnosing deteriorating unit economics or recognizing weaknesses in a pricing strategy.

2. Will we need that capability again?

A founder should understand the economics behind a financial model. They don’t necessarily need to know how to format every spreadsheet.

3. Can we automate the execution while preserving the learning?

Often, we can.

AI can assemble market research while a manager develops the initial hypothesis. It can generate financial scenarios while a founder evaluates their assumptions. It can produce campaign variations while a creative director determines which direction makes strategic sense.

The exercise helps us decide how much of the process should remain with the person responsible for the work.

Illustration: Where to preserve friction

Type of work	Recommended approach
Repetitive administration	Automate
Data collection and preparation	Automate with appropriate checks
Learning an important new capability	Use AI to explain, challenge, and provide feedback
Strategic diagnosis	Develop an initial hypothesis before reviewing AI recommendations
High-consequence decisions	Use AI for analysis while preserving informed judgment and accountability

The aim is to preserve the parts of a workflow that develop capabilities the organization will continue to need.

What this Means for Organizations

Imagine a company introducing AI across its strategy, marketing, and operations teams.

Reports become faster to prepare, campaign production increases, and research takes considerably less time. By conventional productivity measures, the implementation looks successful.

Now consider the junior analyst who previously developed judgment by investigating difficult business problems, or the marketing manager who learned positioning by testing competing hypotheses.

If those learning opportunities disappear, the company needs to find other ways to develop the same capabilities. Otherwise, it may find itself with increasingly sophisticated output but too few people capable of independently evaluating it. 

I think this deserves a place in how executives assess AI adoption.

Alongside the time and costs saved, organizations should examine whether their people are becoming better at the decisions their roles require. That could mean introducing independent hypothesis development into strategic reviews, asking managers to defend the assumptions behind recommendations, or periodically testing how well employees apply what they’ve learned to unfamiliar problems.

These practices needn’t slow down everyday operations. They can be built into existing management and review processes.

Some Things Should Stay Difficult

I have no desire to return to spending hours gathering information that AI can retrieve in seconds. The productivity gains are too valuable to ignore.

But I do think we need to pay closer attention to what people learn while doing difficult work.

A frustrating task may be poorly designed. It may also be exposing someone to a problem they haven’t yet learned how to solve. Those experiences can look remarkably similar in the moment, even though their long-term value is quite different.

As more intellectual work becomes easier to automate, deciding which capabilities people should continue developing will become an increasingly important management responsibility.

Because the value of a task sometimes lies in what it prepares someone to do tomorrow.

Remove the friction that slows the work. Preserve the friction that makes the worker better.

---