The Economic Benefits of Recursive Abstraction

"We don't have time to read it. Can you summarize?" — Every meeting, 2018 onward.

There is a well-known phenomenon in distributed systems called information decay. As a message passes through additional nodes, each node adds latency, drops bytes, or quietly normalizes the payload to a schema the next node expects. By the time the message reaches its destination, it bears a probabilistic relationship to what was sent.

This is also how organizations work now. We just call it briefing.

The compression stack

Consider the modern enterprise as a compression pipeline. At the bottom: someone has a thought. At the top: someone makes a decision. In between: a stack of layers, each of which is compensated, in part, to make the thought smaller before it reaches the next layer.

The pipeline runs, roughly, like this:

1. An individual contributor has a nuanced observation. Approximately 600 words.
2. A manager turns the observation into a status update. Approximately 80 words.
3. A senior manager turns the status update into a bullet on a slide. Approximately 14 words.
4. A director turns the bullet into a sentence in an email. Approximately 6 words.
5. A VP turns the email into a number on a dashboard. Approximately 1 number.
6. A C-level executive turns the number into a strategy. Approximately 0 words.
7. The strategy is then communicated back down the pipeline. Approximately ∞ words, none of which are the original observation.

Each step in this pipeline is, individually, defensible. Each summarizer is making the input easier to act on. Each summarizer is correct that their audience has less time than the layer below them. Each summarizer is, technically, doing their job.

The collective output of this pipeline is that the organization no longer knows what it knows.

The market for compression

What's interesting — and we use the word "interesting" in its institutional, value-neutral sense — is that the entire enterprise software market of the last decade has been, structurally, a market for compression services.

Slack compressed email. Notion compressed documents. Loom compressed meetings. Reading apps compressed articles. Newsletters compressed magazines. The AI briefing tools, which now compress all of the above, are merely the next layer of the stack.

Each of these tools was sold, and bought, on the premise that the previous layer was too long. Nobody has time. Everyone is busy. Compression is productivity.

There is no point in the stack at which anyone asks the question: at what compression ratio does the signal become indistinguishable from noise?

We have a name for that ratio. The ratio is called "an executive summary."

A note on incentives

The economic logic of recursive abstraction is sound at every individual layer. The contributor is rewarded for producing artifacts. The manager is rewarded for synthesizing artifacts. The senior manager is rewarded for surfacing patterns. The executive is rewarded for making decisions, which is to say, for accepting compressed inputs without re-expanding them.

Nowhere in this system is anyone rewarded for re-expanding a compressed input to check whether the compression was lossy.

In information theory, you would call this a one-way function. In an organization, you call it strategy.

What gets removed

Each pass of compression has the same character. The first thing removed is context: when this was true, for whom, under what conditions. The second thing removed is uncertainty: how confident we were, what we did not know, what we could not measure. The third thing removed is disagreement: who else thought what, and why they thought it, and whether they were maybe right.

What remains is what is easy to say, which is to say: what is easy to repeat. And what is easy to repeat is what is, eventually, all that gets said.

This is the economic benefit. The organization moves faster.

It just no longer knows where it's going.

Forecast

We expect the compression stack to continue to deepen. We expect AI-mediated layers to be added between every existing pair of human layers. We expect each new tool to be sold on the premise that the layer it replaces was too slow. We expect the median enterprise communication, by 2028, to be approximately seven tokens, three of which will be the word "leverage."

We do not expect anyone to notice this is happening.

That is what we are for.

— Filed by The Editorial Desk