Origin
As an intern at x firm, I did something unremarkable: I documented my workflows in a markdown file. What I actually did, across which tools, in what order, and why. The company built an entire program around that file. Not because the information was novel — because no one had ever made it legible. The knowledge was always there. It was just trapped in the people who held it.
01
The middleware layer
Most of a software org doesn't build software but rather It translates. Product managers translate customer signal into engineering requirements. Analysts translate output into executive legibility. Program managers translate timelines across teams. This is the coordination middleware — an entire class of roles that exist because the people who build and the people who decide can't efficiently share context. The layer is a structural sickness. It exists because information asymmetry is expensive and humans are bad at bridging it at scale.
02
The asymmetry is collapsing
Consultancy is the clearest proof. The entire business model is an arbitrage on information asymmetry. If a company does not know something, it pays someone else who does. AI reduces the cost of that arbitrage to zero along with eliminating the conditions that create it. When a model can passively synthesize what engineers actually do across every tool they touch, the question isn't "who translates this?" — it's "why would anyone need to?" The same collapse is coming for every role whose primary value is moving context from one format to another.
03
Engineers + leadership. That's the org.
The companies that move fastest will be the ones that stop hiring translators and start investing in infrastructure that makes translation unnecessary. The end state is an org with two layers: people who build and people who decide. AI absorbs everything in between. Individual contributors will not be replaced, but by making the connective tissue between builder and decider so efficient that a human in that role becomes overhead.
So what
This is an architectural prediction. Layoffs are only part of it. The orgs that figure out AI-native structure first will operate at a fundamentally different clock speed than those still paying humans to carry context across hallways. The transition won't be theoretical — it'll show up in headcount ratios, in decision latency, in how fast a company can go from "we should build this" to shipping it.
That's what I'm building toward. Furlough is the infrastructure layer for companies that see this coming and want to move before their competitors do.