The Agentic Company Already Exists.

There is a table circulating in AI circles that maps agent terminology to organisational terminology. Skills are employees. The resolver is an org chart. Filing rules are internal process. Check-resolvable is audit and compliance. Trigger evals are performance reviews.

The table is presented as an analogy. It is not an analogy. It is a description.

Notice the two empty rows at the bottom. They matter more than the five above them. We will come back to those.

The costume

The AI community has spent the last three years reinventing management science from first principles, without realising that management science already existed. Routing logic is what operations managers call escalation policy. Skill registries are what HR calls a competency matrix. The "resolver" is the org chart. The question "can the system do what it claims?" is what auditors have been asking since the concept of a limited company.

This matters because of what it reveals about who is building these systems and who is not. The people designing agentic architectures are, overwhelmingly, software engineers. They bring excellent intuitions about modularity, composability, and abstraction. They bring almost no intuition about what happens when you deploy a system of autonomous workers into an environment where outcomes depend on coordination, escalation, and institutional memory.

The absent expertise

The people who do have that intuition are not in the room.

Operations directors. Manufacturing engineers. Programme managers who have spent careers designing systems where autonomous agents (called "employees") collaborate under imperfect information, partial authority, and real consequences for failure. These people know what a bad org chart costs. They know what happens when escalation logic breaks. They know that the hardest problem is not routing the work to the right team but knowing whether the outcome was good after the team responded.

The pattern is consistent. Every time a new technology creates a new organisational form, the people who understand the predecessor form are excluded from building the successor. Manufacturing went through this with automation. Finance went through it with algorithmic trading. Healthcare is going through it with clinical AI. The technical architecture is the easy part. The organisational architecture is where systems live or die.

The missing rows

Back to those two empty cells in the table.

The first missing row is feedback loops. The table includes trigger evals, mapped to performance reviews, which check whether the right team responded. But that is not the same as checking whether the response worked. Did the customer get helped? Did the decision lead to a good outcome? Did the system learn from what happened?

The gap between process compliance and outcome measurement is where most organisations, and most agent systems, quietly fail. They optimise for routing accuracy and call it performance. They measure whether the ticket went to the right queue, not whether the problem got solved.

The second missing row is institutional memory. Filing rules say where information lives. Nothing in the table addresses how information accumulates, how it ages, how it gets interpreted by someone who understands the context well enough to act on it. In organisations, this is the person who has been there fifteen years and carries the unwritten rules in their head. In agent systems, it is usually nowhere.

Both gaps share a root cause. They are the difference between managing a process and managing for an outcome. Processes are designable. Outcomes require something harder: judgement shaped by experience, applied in real time, under conditions the process designer did not anticipate.

That is what the smiðr carried. That is what the manufacturing engineer carries. And that is what agent architectures do not yet have a row for.

The recognition

A new technology inherits the structural problems of the systems it replaces, plus the additional problem that nobody in the new system recognises the old problems as their own. The vocabulary changes. The failure modes do not.

The agentic company does not need to be invented. It needs to be recognised. The org chart is the resolver. The competency matrix is the skill registry. The escalation policy is the routing logic. The institutional knowledge is the thing everyone forgets to build until the system breaks in a way that was obvious to anyone who had been there long enough.

The question is not whether agent systems need organisational design. They are organisational design. The question is whether the people who understand organisational design will be in the room before the architecture is set, or after the first expensive failure.

History suggests after. But it doesn't have to.

The product

Kaipability is built this way. Not as theory. As operating architecture.

Every capability the firm delivers is a skill with a trigger condition, a scope boundary, and a quality checklist. The resolver is a human-AI centaur: a chartered engineer and an AI system that carries institutional memory, voice calibration, and methodology between engagements. The filing rules are codified. The feedback loops are built. The audit function exists as a meta-skill that checks whether the other skills can deliver what they claim.

This is not a consultancy that uses AI tools. It is an AI-native firm where the organisational architecture and the agentic architecture are the same thing. The skills are the employees. The methodology is the org chart. The accumulated judgement is the institutional memory.

Your factory is your product. That has always been the founding insight. Applied reflexively, it means the architecture of Kaipability itself, the way conviction, capital understanding, and capability are coupled inside a single operating system, is what scales. Not the consultant. Not the slides. The system.

The agentic company does not need to be invented. We are already building it.