A Proposed Operating Model for Hybrid Teams of People and AI in Fractal Symbiosis

Agenda, James

Working paper · Version 0.1 (beta)

A Proposed Operating Model for Hybrid Teams of People and AI in Fractal Symbiosis

James Agenda October 1, 2026 · Independent researcher · 12 min read

Research in progress. This is a working paper about a model under active research and daily testing. It is shared early, in beta, so others can use it and help improve it. Expect changes between versions.

Abstract

Organizations are starting to work with AI agents, but most current proposals describe large enterprises seen from the top and treat the agent as one more employee. This paper proposes an operating model built from the smallest unit up: one person and one AI symbiont, a dedicated agent with full access to that person's machine. The model splits work into three kinds of structure (ventures, processes and process groups), separated by a single test: who pulls the change. Each structure carries a DNA written only by the human and a DNA-X written only by the symbiont, which the symbiont loads before every session. The same structure repeats at every layer, from a single process to a whole company or a whole life, so the model is fractal. We describe the model, its biological inspiration, its relation to the Viable System Model, Holacracy and organizational ambidexterity, and a first implementation on one real case. The model is published in beta, with open files that an AI agent can read and implement.

Keywords: operating model; organizational design; human-AI teams; agentic organization; AI agents; fractal organization; symbiosis; governance; decision rights; knowledge management.

1. Introduction

In Department of Artificial Resources I argued that every company will need a department to manage AI agents, and that an AI-first company has to redesign its processes and structure its knowledge so agents can consume it. The book stopped there: it said the redesign was needed, but did not propose one.

Since then I have faced the same challenge every day: how do you actually organize work when part of the team is human and part of it is AI? It is still an open problem, and many people are testing answers. This paper is my current answer, in beta. It is an operating model, which means it goes beyond organizational design: it covers organizational design (how work is split), decision rights (who decides what, human or AI), governance (the rules and how they change) and knowledge management (what the AI always carries, what it looks up, and where each thing lives).

2. Working assumption

Each person has a dedicated machine and a symbiont: an AI agent, an LLM, or any other artificial intelligence or artificial resource, with full access to that machine. The person talks to the symbiont through conversation. The symbiont is the one who reads the model's documents and puts them into practice. The scarce resources are the human's time, attention, energy and physical presence; the symbiont's capacity is close to unlimited. The model puts the scarce resources where they matter most and hands everything else to the symbiont.

3. The model

3.1 Three kinds of structure

Structure	Exists to	How it changes
Venture	Seek. It has a goal.	Proactively. Changes are pursued on purpose, from early hypothesis testing to a stable venture on autopilot. Reviewed at least quarterly.
Process	Maintain.	Reactively. Only when something outside asks: it broke, or a good opportunity appeared. It can run for years untouched.
Process group	Group processes that share rules or knowledge.	Follows its processes.

The test that separates them is a single question: who pulls the change? If it comes from outside, it is a process. If we go after it on our own initiative, it is a venture. This mirrors the exploration and exploitation split described by March (1991).

3.2 DNA and DNA-X

Every structure carries two documents. The DNA is written only by the human: the objective in one line, the goal (for ventures) and any rules the human wants to enforce. The DNA-X is written only by the symbiont: what it learned and needs to carry every time it works on that structure, capped at one page. The DNA always wins over the DNA-X. Everything else the symbiont writes is a note, owned by one structure and read only when needed.

The inspiration is biological. The DNA-X behaves like a plasmid: extra genetic material a bacterium picks up from its environment and keeps because it helps it survive. The symbiont acquires knowledge from the work and keeps only what proves useful.

3.3 Hierarchy and fractal recursion

Documents follow a strict order of authority: the constitution, the symbiont's DNA, the DNA of each structure from the top down, then the DNA-X files, then notes. Rules flow down: what a structure forbids is forbidden in every structure below it. The symbiont loads the chain of DNAs from the top structure to the one it is working on, before acting.

The same shape repeats at every layer. For one person, the top structure may be their whole life. For a department head, it is the department. A delegated venture is where the tree passes to another person and their symbiont, possibly on another machine. The model therefore works for one person and one agent, and for a thousand people and a thousand agents.

3.4 Processes, metrics and alarms

A process is either automatic (runs with AI and automation; the human only steps in when it fails) or human-involving (the human takes part every time, following a written checklist). Each process has an objective, a cadence, a definition of success, a few metrics it updates itself and, if the human wants, alarms. A dashboard generated from the files shows every structure: dark when running well, white when not yet built, red when it needs the human.

4. Related work

The model is closest to Beer's Viable System Model, in which every viable organization contains the same functions recursively at every level. Our constitution and DNA map to identity, ventures to intelligence and adaptation, the dashboard and the allocation of human resources to control, an independent checker to audit, processes to operations, and red alarms to Beer's algedonic signals. The coordination function between sibling units is not yet explicit in our model.

Holacracy also runs on a written constitution and nested circles. Organizational ambidexterity (March, 1991) separates exploration from exploitation, as ventures and processes do here. Current consulting work describes the agentic organization (McKinsey) and agentic operating models for governing autonomous AI at scale (California Management Review), with humans moving from approving every action to setting the boundaries agents act within. Those proposals start from the large enterprise. This one starts from the individual and composes upward.

In evolutionary computation, Watson and Pollack showed that symbiotic composition helps adaptation in landscapes that are hard at every scale, which is the setting a fractal organization faces.

What we have not found elsewhere is the combination of: a structure that repeats at every layer, documents that the agent itself loads and executes before each session, a strict split of authorship between human and AI, and the person with their symbiont as the basic unit.

5. Implementation

The model runs on one real case: one person, one symbiont, one dedicated Linux machine, with personal life, two companies and partner projects as structures. Delegated ventures are handled by other symbionts on other machines. The constitution, protocols and dashboard evolved in daily use; the three-structure split and the "who pulls the change" test were defined on October 1, 2026.

6. Limitations

This is a single case (n = 1), observed by its own author. The model has not yet been tested with multiple people, with teams, or with symbionts other than the one used here. Coordination between sibling structures and the propagation of rule changes across machines are open problems. It is a beta and will change.

7. Availability

The files are open. They are written for an AI symbiont to read and implement. Ask yours:

Read fractal-symbiosis-by-james-agenda.txt from this page, study it, and propose how to adopt it here.

Download Fractal Symbiosis (.txt) Download the package (.zip)

Contents: README-FIRST.txt, CONSTITUTION.md, PROTOCOLS.md and blank DNA and DNA-X templates for the symbiont, for structures and for processes.

References

Agenda, J. Department of Artificial Resources: The New Management in the Age of AI Agents. jamesagenda.com/en/book/dra
Beer, S. (1972). Brain of the Firm. Allen Lane.
March, J. G. (1991). Exploration and exploitation in organizational learning. Organization Science, 2(1), 71-87.
Robertson, B. J. (2015). Holacracy: The New Management System for a Rapidly Changing World. Henry Holt.
McKinsey & Company (2025). The agentic organization: Contours of the next paradigm for the AI era. mckinsey.com
California Management Review (2026). Governing the agentic enterprise: A new operating model for autonomous AI at scale. cmr.berkeley.edu
Watson, R. A. & Pollack, J. B. (2003). A computational model of symbiotic composition in evolutionary transitions. BioSystems, 69(2-3), 187-209.