Definition

Platform compression

Platform compression is the collapse of the middle ground between infrastructure owners, platform owners and feature owners. AI lets the first two absorb features directly, which leaves the third without defensible ground.

A note on the words first, because they are already taken. In engineering, "compression" on a platform usually means model compression, response compression, or quantisation. This is not that. The subject here is the compression of the platform cycle itself: the shortening of the interval between a capability becoming technically possible and it becoming a free feature of somebody else's platform. Where that interval used to run in years, it now runs in quarters.

The distinction from the existing literature on platform economics matters too. Wikipedia, Belleflamme and Peitz, Parker and Van Alstyne and the rest describe platform structure: network effects, multi-sided markets, who captures value and why. Platform compression describes a rate. It is a claim about speed, not about shape, and it is the speed that has changed.

The three rule shifts

1. Efficiency over headcount

AI replaces junior expertise at scale, and the saving does not sit in the margin. It funds platform expansion. The organisation that cuts the junior tier and books the saving has taken the cost and skipped the strategy. The organisation that cuts the junior tier and spends the saving on capability others can build on has done the thing the saving was for. Same headcount number, opposite outcome, and the difference is invisible in the quarter it happens.

2. Value over volume

Platform owners monetise the signals that other people generate. The metric that used to matter was throughput: calls served, transactions processed, seats filled. The metric that matters now is what the traffic reveals and who is allowed to read it. Access to data stops being a side-effect of the service and becomes a priced product in its own right.

3. Platforms over features

Standalone capabilities become free functions of an operating system. This is the oldest of the three and the one people still misread as bad luck. A feature absorbed into a platform was never a business; it was a gap in someone else's roadmap, monetised for as long as the gap lasted. What changed is how long that is.

A worked example

In 2018 I published a four-part framework arguing that platforms would concentrate power and that APIs would democratise expertise. Fewer than two thousand people read it. The direction was right. The velocity was wrong by roughly a decade, and being right about direction while wrong about speed is functionally the same as being wrong.

The framework itself is now a documentation page inside a hyperscaler's cloud product. Not because anyone took it, but because that is what happens to a description of how something works once the platform that it describes decides the description is part of the product. The expertise did not disappear. Its scarcity did. That is compression at the scale of one person, and it is the same mechanism that operates on a company.

How to tell whether it is happening to you

The diagnostic is not market share and it is not growth. It is whether the thing you charge for would survive the platform beneath you shipping it as a checkbox. If the honest answer is that it would not, the question is only how many quarters you have, and the answer to that question has been getting smaller since 2023.