The thesis

AI is a structural shift. Productivity is a side effect.

A recursive, self-improving company is one where every meaningful signal, decision, action, and outcome becomes machine-readable, verifiable, and reusable in the next cycle of work.

The problem

The destination is increasingly clear. Building it remains the open question.

Every organizational innovation since antiquity solved one recurring problem: how to move information across a large group of people when human span of control is limited.

The old technology

Hierarchy

Named individuals relay information up and down a chain. Every layer adds latency, distortion, and politics.
The new technology

The intelligence layer

A continuously updated model of the organization routes information with no span-of-control constraint — parallel, queryable, and improving overnight.
The framing

Bolt it on, or build it in.

Most companies

Bolt AI onto existing workflows

Treat AI as a productivity tool that accelerates existing teams. The org chart, workflows, and systems of record stay intact. Result: a smarter dashboard.
Our view

Build a client-owned System of Intelligence

Redesign the company so it senses, orients, acts, verifies, and learns — without manual coordination at every step. Result: a company that compounds.
The gap

Most companies are building AI workbenches. The intelligence layer stays empty.

The prize is the orchestration layer — the reasoning above the database.

An AI workbench

Tools wired into a CRM

  • · A CRM with agents wired in
  • · Slack, email, Gong, enrichment connected
  • · A prioritized feed of AI suggestions
  • · Copilots that make people faster
A System of Intelligence

Sensemaking, governance, learning

  • · Stable entities, orientation, causal reading
  • · Task-node contracts, routing, evidence plane
  • · Evals, proofs, and memory that compounds
  • · The advantage continues while the company sleeps
The implication

If your AI plan depends on the frontier model staying where it is, you don't have a plan.

Build value above the model. The model provides inference. The moat is context, workflow, policy, evidence, memory, and learning.

Commodity layerDurable value layer
Frontier modelDecision environment
PromptingEntity + context layer
Agent actionsTask-node contracts
Output qualityEvals and proofs
Tool useGoverned execution
Conversation historyInstitutional memory
Most AI work is brittle, unstructured, and model-vulnerable. Differentiation collapses as models get better, faster, and cheaper.
What we see in the market