Technology is becoming easier to build.Competitive advantage isn’t.

For four decades, technology capability was scarce because building it was expensive. It took large teams, specialised expertise, infrastructure, time and capital. That cost was the moat, whether or not anyone called it one — in software, and equally in engineering, planning, analysis and administration.

AI removes part of that scarcity. Not evenly, and not everywhere at once. But the floor under too expensive to copy is falling, and it is falling faster than most three-year plans assume.

The value does not disappear when creation gets cheaper. It moves — toward the things a competitor cannot assemble by building. So the question in front of a board or an investment committee changes.

What remains defensible when others can build it too?

AI Compression is the reduction in cost and time required to perform an economic activity through artificial intelligence. It reaches software development, engineering, analysis, administration, support, documentation, planning, design, quality work and parts of customer interaction.

A cost reduction, though, is not a margin until someone fails to compete it away. If an activity becomes fifty percent cheaper, the benefit can land almost anywhere: with the company, with its customers through lower prices, with employees, with a competitor who never carried the old cost base, or with the platform that supplies the capability.

If engineering productivity doubles for every competitor at once, the result may be higher margins — or lower barriers to entry, or faster feature expectations, or lower prices, or some combination. Which one it is depends on the structure of the market, not on the technology.

Productivity improvement is not automatically competitive advantage.

When creating technology gets cheaper, organisations create more of it: more applications, automations, workflows, agents, integrations, internal tools and models. That is mostly good. It is also how dependency accumulates.

More systems built faster, by more people, in more places, means more hidden dependencies, more fragmented ownership, more technology that no function formally owns, thinner documentation, wider security exposure and more processes resting on something nobody has examined.

The constraint moves. It used to be capability. It becomes governance and judgment.

The question shifts from “can we build it?” to “should we build it, and can we safely depend on it?”

Excel made computing accessible to people who were not programmers. That created enormous value, and it is still creating it. It also let temporary solutions become permanent infrastructure.

The sequence is familiar. Someone needs a calculation. The spreadsheet becomes a workflow. The workflow becomes a business process. The process becomes operationally critical. Then the author leaves, several versions circulate, the formulas are no longer fully understood, and manual workarounds accumulate around something that was never designed to carry weight.

Excel was not the problem. The problem was that a cheap tool became responsible for an expensive process, and nobody noticed the moment it happened.

AI, low-code and agentic software may reproduce that pattern at far greater speed. A department can now produce an application, an approval workflow, a planning tool or an agent in hours. That the thing can be built quickly answers none of the questions that decide whether it should be depended on.

The danger is not that people can build. The danger is forgetting when what they built became critical.

Most technology decisions are still argued on the cost of creation, because that is the number a proposal contains. It is also the number AI changes most, and the one that matters least once a system is in production.