Can We Become More Efficient and More Fragile at the Same Time?

Can We Become More Efficient and More Fragile at the Same Time?

Most business technology arrives with a simple promise: do more with less effort.

Can We Become More Efficient and More Fragile at the Same Time?
Most business technology arrives with a simple promise: do more with less effort.

Generative AI makes that promise unusually convincing. A small company can draft documents, analyse information, produce code, translate, summarise, research and create material that once required more people, more time or more specialised knowledge. For many tasks the gain is not theoretical. It is already visible in ordinary work.

This is why I find some of the debate around AI slightly misplaced. The interesting question is not whether companies will use it. If a tool makes useful work cheaper and faster, businesses have a strong reason to adopt it. The interesting question comes afterwards.

What happens to the company when a capability that used to live partly inside people begins to live inside a service the company does not control?

Efficiency and resilience are often treated as if they naturally travel together. Remove duplication, automate the repetitive work, concentrate on core activities, use the best external tools. Each decision can improve performance. It can also remove slack, alternatives and knowledge.

The capability can disappear before the tool does

The obvious dependency is technical. A provider may change price, limits, product design, access conditions or availability. A model that works well today may be replaced. An API can change. A feature can disappear into a higher-priced plan. Businesses already know this kind of vendor risk.

AI adds another layer because the external tool may not simply store information or execute a process. It may become part of how people think through the work.

Suppose a team uses AI to draft almost every proposal. At first, experienced people use it as an accelerator. They know what a good proposal looks like, recognise weak reasoning and correct the output. Over time, new employees learn the workflow differently. The first draft always comes from the system. Templates disappear because nobody needs them. Examples are no longer maintained. The senior person who knew why certain arguments mattered leaves.

The AI service is still working perfectly. The internal capability has nevertheless changed.

This is more difficult to measure than an outage. There is no red light showing that the organisation has lost part of its independent ability to perform the task. Productivity may actually continue improving while the dependency deepens.

That is the uncomfortable possibility: the company becomes better at producing the output and worse at producing it without the tool.

Redundancy is inefficient until the day it is useful

Management has spent decades learning to remove duplication. Usually for good reason. Two systems doing the same job cost money. Manual fallbacks are slower. People maintaining skills that software can provide may look wasteful. Resilience often contains exactly these apparently inefficient things.

A second supplier. Documentation nobody needs on a normal Tuesday. A person who still understands the underlying process. Data stored in a format that can move. A manual procedure that is slower but works. Enough knowledge to recognise when the automated answer is wrong.

The objective is not to preserve every old method forever. That would turn resilience into nostalgia and make organisations expensive for no good reason.

The useful question is proportional: which capabilities would hurt enough if lost that we should preserve some ability to replace, reconstruct or operate them differently?

AI makes this question urgent because adoption can happen faster than organisational memory changes. A tool can enter hundreds of small workflows without any single moment at which somebody declares, “This is now critical infrastructure.” The dependency is built one useful shortcut at a time.

The same logic applies to judgement. If people use an AI system to challenge their thinking, the tool can strengthen the work. If they stop being able to recognise when its answer is shallow, incomplete or simply wrong, the organisation has outsourced something more important than drafting.

When I started working on Dependent Intelligence, this was the contradiction I found hardest to ignore. I am not interested in an anti-AI argument. The technology is too useful for that to be convincing. The risk comes from success.

The better these systems become, the more rational it is to use them. The more we use them, the more carefully we need to understand what we are allowing to disappear around them. A company can absolutely become more efficient and more fragile at the same time. The difficult part is that the efficiency appears first.


Related book

Dependent Intelligence How Businesses Are Building Critical Capabilities on AI They Don’t Control

A business-focused examination of what happens when AI becomes critical to workflows, knowledge and decision-making while the underlying models, infrastructure, pricing and product direction remain outside the company’s control.

Available on Amazon → [link pending]

Scroll to Top