By late 2025 the thesis was formed. The harness layer was being absorbed. We published it in early 2026. The market confirmed it within weeks.
The middleware stack built to orchestrate, coordinate, and govern large language models was structurally exposed — not because the vendors who built it were wrong, but because the providers who owned the models would inevitably absorb its function as a native primitive. We called it the Great Compression. We said the harness is the moat, and the provider knows it.
Most of the people who read it agreed with the dynamic but thought the timeline was years away.
Claude Managed Agents shipped in April 2026 with memory, dreaming, outcomes, and multiagent orchestration. In a single release, the provider absorbed what the ecosystem had spent three years building under two names — scaffolding first, then harness — neither of which it fully earned. The compression was not gradual. It was structural — and then it was done.
But the Great Compression only named the dynamic. It did not name the destination.
This series does.
What the compression revealed
When the harness layer compressed, it exposed something the middleware stack had been obscuring: the full architectural picture of what the agentic era actually requires.
The harness was not just an orchestration layer. It was a compensation layer. It compensated for stateless models by managing context. It compensated for ungoverned LLM use by adding evaluation scaffolding. It compensated for the absence of a living data substrate by building retrieval pipelines on top of passive datalakes. It compensated for cognitive-era audit frameworks by logging what it could and hoping regulators would not look too closely.
When the provider absorbed the harness, those compensations became native capabilities. And the architecture underneath them — the enterprise systems they were compensating around — became visible in full relief.
What that architecture revealed is the subject of this series.
The systems of record that enterprise operations were built on were never designed for agents. They were designed for humans. Every field, every workflow, every approval screen encoded a human cognitive model that autonomous agents cannot use and were never meant to use.
The datalakes that were supposed to unify enterprise intelligence were passive storage systems organized by the humans who built them — not by the decisions agents need to make. The intelligence was never in the lake. It was in the analyst who knew which question to ask.
The audit frameworks that were supposed to govern AI systems were written for a world of discrete human decisions in pre-defined workflows. Agentic actions are runtime-determined and compound on themselves. There is no pre-defined workflow to audit against.
And the infrastructure bets placed by the largest capital allocators in technology — hundreds of billions in data centers, middleware companies, AI-native application replacements — were optimized for a world where the model was a commodity and the value lived in the infrastructure around it.
None of those assumptions survived contact with the refinement layer.
What the Displacement names
The Great Compression described a structural dynamic. The Displacement describes its architectural consequence. Five structural shifts define the agentic enterprise architecture that emerges on the other side of the compression.
Why this is the sequel
The Great Compression was about the structural dynamic — the gravitational force pulling the middleware layer into the provider stack.
The Displacement is about what that force produces on the other side. Not the compression itself, but the architecture it reveals when the scaffolding falls away.
These six posts are not independent arguments. They are a single continuous argument in six movements. Each post advances one structural shift. Each shift builds on the one before it. By the time the sixth post lands, the full architecture of the agentic enterprise is on the table — not as a prediction, but as a description of what is already being built, already being funded, and already being decided in enterprise architecture conversations happening right now without a complete framework to navigate them.
This series is that framework.
It will make some people uncomfortable. The SOR vendors whose position is being precisely described. The datalake companies whose thesis is being precisely challenged. The venture firms whose portfolio companies sit in the compression zone between the SOR they were funded to displace and the provider substrate that is compressing them from above. The enterprises that have been accumulating substrate dependencies without a deliberate architecture and are about to discover what that means.
We are not writing to make anyone uncomfortable. We are writing because the architecture is what it is — and the practitioners who understand it clearly will make better decisions than the ones who are still operating on assumptions the compression already invalidated.
The Great Compression named the dynamic.
The Displacement names where it leads.
