Thesis
Building the recursive, self-improving company.
A company that senses, orients, acts, verifies, learns, and improves on its own. The destination is agreed by most people who think about this seriously. What does not exist yet is a way of actually getting there.
The shift
Value is migrating upward.
Systems of record survive. They become inputs to the layer that orients, decides, acts, verifies, and learns. In the software era, gravity came from accumulating data in one place, because a person could only look in one place. In the AI era it comes from orchestration, because an agent can read many places at once.
Agent and user layer
Takes action, queries, decides
Where a person or an agent actually does the work.
System of intelligence
Orchestrates, synthesizes, routes, learns
The layer that holds context, enforces policy, keeps the record, and improves from it.
Systems of record
CRM, ERP, HRIS
Still running, still authoritative for their own data, increasingly consumed at the API layer.
Gravity moves upward
The company becomes a queryable model of strategy, work, decisions, and customer context.
Jack Dorsey
The platform
A client-owned system of intelligence.
Owned by the company rather than rented from whoever sold it. Three parts, and each one is useless on its own.
Sensemaking
The continuously updated world model
Stable entities, orientation, and causal reading. A working picture of what changed and why.
Governed execution
Action under contract
Task-node contracts and model routing, with an evidence plane beneath every action the system takes.
Verified learning
Memory that compounds
Evals and proofs, with memory that carries forward. A workflow built this way keeps improving while the company sleeps.
Assistance makes individuals faster and leaves the company unchanged. With all three in place, what the company learns accrues to the company rather than to whoever happened to be using the tool that week.
The ladder
Queryable, then proactive, then recursive.
Most companies stop at queryable. The advantage compounds after proactive detection and verified learning, which is the part that requires a record.
01
The queryable company
You can ask what happened. Work, data, decisions, and context are accessible to AI rather than trapped in tools and inboxes.
02
The proactive company
The system surfaces what matters. Important changes are detected before a human thinks to ask about them.
03
The recursive company
The system acts, verifies, learns, and improves. A verified outcome updates memory and policy, and changes what runs next time.
How the two readings fit
One ladder is about capability. The other is about governance.
One ladder runs from queryable to proactive to recursive, and it describes what a system can do. The other runs from individual use to governed use to self-improving, and it describes how a company has organized itself. They are different axes. A company can sit high on one and low on the other, which is the source of most confusion in this market.
Capability
A company can be proactive on paper because a vendor shipped detection. That capability arrived from outside and can leave the same way.
Governance
A company only becomes recursive if it keeps a record of what worked. The record is what makes improvement accrue rather than evaporate, and no vendor supplies it.
The ladder stalls at queryable for so many companies because of this split. The first two rungs can be purchased. The third one cannot, because it depends on evidence the company itself produces.
The question
If you were building your company today, knowing what is possible, would you build it the way it is organized right now?
For most companies the answer is no. The gap between that answer and the current org chart is the work.
We build the recursive, self-improving company with the people already inside it.