Ration the Shutter

For most of history, good answers were not free. You had to know something, find someone who did, pay for expertise, search through documents, make a calculation, run a test, or spend enough time thinking that an answer eventually revealed itself.

Artificial intelligence is driving that cost down with remarkable speed. A small business can now ask for an analysis of a contract, a comparison of pricing models, or a plan for entering a new market and receive something useful in seconds. Not flawless or infallible, but often good enough to move the conversation forward. That may be the more important AI story: not that machines are becoming brilliant, but that reasonable competence is becoming cheap.

Economists pay attention when the price of something important collapses, because people behave differently when scarcity disappears. Digital photography did not simply make photographs cheaper. It changed how many we take, what we photograph and what a photograph means. We stopped rationing the shutter. AI may do something similar to answers.

But AI arrives with an unusual complication: almost every advantage it creates produces a tension somewhere else. Business leaders see an opportunity to move quickly; employees may see the same movement as a threat to their livelihood. An owner sees a tool that allows ten people to accomplish what once required fifteen. The ten people reasonably wonder what happens when it requires eight. A company sees an opportunity to capture the knowledge of its most experienced employees. Those employees may wonder whether documenting everything they know makes them more valuable or easier to replace.

The debate around AI is often framed as optimism versus pessimism, adoption versus resistance. In practice, the tensions run straight through individual organisations. The same technology can make a business stronger while making some of the people inside it feel less secure. Imagine asking employees to identify the most repetitive parts of their jobs. From the employer's perspective, it is sensible; the people closest to the work know where the friction is. From the employee's perspective, it can sound suspiciously like being asked to prepare a list of reasons their job should no longer exist.

The information the business most needs may therefore be held by the people with the least incentive to provide it. There is a paradox here: AI becomes more useful when an organisation understands how work really happens, but much of that understanding exists inside people, in shortcuts, exceptions, relationships and accumulated experience. Accessing it requires trust precisely when the technology itself may be undermining that trust.

As the answers get better, another tension appears. When an organisation can generate ten possible strategies as easily as it once produced one, producing options becomes easy; choosing among them becomes harder. AI can dramatically reduce the cost of reaching a decision. It does nothing to reduce the cost of being wrong.

The answer is cheap. The consequence is not.

A recommendation to stop carrying an underperforming product may make perfect sense in a spreadsheet, while the owner knows people drive across town specifically for it and buy three other things while they are there. A staffing model may identify two employees as doing essentially the same work, while their colleagues know one of them is the person everyone calls when something unusual happens. A supplier may appear more expensive than its competitors until something breaks late on a Friday afternoon and the reason suddenly becomes obvious.

Businesses are full of information like this. Context is the accumulation of things that seem irrelevant until suddenly they are not.

The better AI becomes, the easier it will be to mistake a complete-looking answer for a complete understanding of the situation. We should also be careful about assuming humans will always possess some category of judgement machines cannot reach. History gives us little reason for that confidence. Tasks we once considered uniquely human have an inconvenient habit of becoming computable.

There is, however, another distinction: AI can participate in a decision, but it cannot be accountable for one. A system can recommend laying off an employee, but it cannot look that person in the eye. It can recommend closing a location, but it does not live in the community after the doors are locked. It can identify the most efficient option, but it has no stake in deciding whether efficiency is the only thing worth preserving.

That does not make its recommendation wrong. One day, its analysis may routinely be better than ours. But advice and accountability are different things. The accountant who gave advice attached a professional reputation to it. The contractor who recommended a solution had to build it. The manager who changed a process lived with the employees affected by it. The business owner faced the customer if the decision went badly. AI begins to separate the recommendation from the person living with its consequences.

A similar tension appears in employment. Much AI adoption naturally targets routine and entry-level work. If software can perform the first draft, basic research or repetitive administrative work in a fraction of the time, removing that work appears to be an obvious gain.

Some work has always performed two jobs. It produces something. And it produces somebody.

The junior accountant checking routine files is also learning what an unusual one looks like. The apprentice doing the simple repair is accumulating the pattern recognition required for the difficult one. The inexperienced employee handling ordinary customer questions is learning how customers behave. Viewed purely as output, this work can look inefficient. But the first hundred ordinary tasks are often the price of creating someone capable of handling the hundred-and-first when it is not ordinary.

Businesses have a legitimate incentive to remove work that technology can perform cheaply. Society still needs some mechanism through which inexperienced people become experienced. A ladder remains perfectly functional without its bottom rung. It simply becomes considerably more useful to the people who are already on it.

For small businesses and communities, these tensions matter beyond individual balance sheets. AI could allow a five-person company to access capabilities once available only to a fifty-person company. It could strengthen independent businesses, preserve knowledge, lower barriers to entrepreneurship and free people from work nobody particularly enjoys doing. At the same time, those gains can alter employment, training, customer relationships and the distribution of knowledge inside organisations.

Both things can be true.

Perhaps that is the more useful way to think about the AI landscape: not as a choice between embracing the future and resisting it, but as a series of bargains.

Speed in exchange for what? Efficiency at the expense of what? Convenience instead of what? Knowledge captured from whom, and for whose benefit?

A business can make a perfectly rational decision and still create consequences beyond the thing it was trying to optimize. We are adopting artificial intelligence because it can make organisations more capable. It would be strange to assume that capability will be the only thing it changes.

As the price of a good answer approaches zero, the important questions increasingly sit around the answer itself: who benefits, who feels threatened, what knowledge disappears, what relationships change, and who remains responsible for what happens next.