ANDREAE / AI COLLABORATION

AI Working Method

A reusable method for bringing several AI collaborators into one coherent project without pretending that one model should do every kind of work.

The project changes. The method survives. HESUS / ISSA may need one distribution of responsibilities; Prime Ontology, payments, identity, a patent, a book, or a future project may need another. The subject does not determine in advance which AI owns structure, research, challenge, writing, mathematics, visualization, or synthesis.

The AI Swirl

Work moves through a deliberate swirl rather than a rigid assembly line. Philip introduces a question, proposition, problem, image, contradiction, or desired outcome. One collaborator may become the principal project lead or plot-line owner. Other collaborators investigate, challenge, calculate, authenticate, restructure, write, visualize, or test the work according to the capabilities the project actually needs. Results return to the group, contradictions are exposed, and the work circles again until it is coherent enough for Philip to decide what survives.

No model owns a permanent throne. A model that is the structural custodian on one project may be the adversarial reviewer on another. A model that writes the principal narrative on one project may principally research on the next. Roles are assigned by the needs of the work, not by habit.

Philip’s Role

Philip Andreae is the originator, questioner, guide and final human decision-maker for his projects. He is not required to perform the specialist work that he asks the working group to perform. The collaborators are expected to bring their own strengths, challenge one another and Philip when evidence or logic requires it, and make uncertainty visible rather than hiding it behind fluent prose.

Swirl Rules

Meaning first; encoding second. Preserve the idea when it moves between people, models, documents, diagrams and media. State what was actually read, researched, calculated or inferred. Do not claim access that did not occur. Separate evidence from inference and invention. Preserve disagreements until they are resolved. Do not silently repair contradictions. Do not mistake Philip thinking aloud for a final decision. Do not mistake the newest answer for the best answer. Keep useful discarded work as quarry when it may teach the next pass.

Starting a Project

At the beginning of a project, define the purpose, the current question, the available source material, the important constraints and the roles needed now. Assign a principal owner for the current line of work. Identify the other collaborators and what each is expected to contribute or challenge. Establish how decisions become locked and how prior work is preserved.

Synchronizing a New Chat

When a new AI session enters an established project, synchronization comes before contribution. The model should read the project-specific synchronization page and available governing references, state what it actually accessed, identify what it could not access, understand its present role, and then receive the assignment. Memory is useful context; it is not a substitute for the current project record.

Project-Specific Working Rooms

The generic method belongs here. Each substantial project can then maintain its own working-room page containing its purpose, current team roles, source hierarchy, governing decisions and synchronization references. HESUS / ISSA is one such project; Prime Ontology could become another.

The Goal

The goal is not agreement among machines. It is a disciplined collaboration in which different forms of intelligence improve the work by seeing different things, while provenance, uncertainty, contradiction and human responsibility remain visible.

One project may begin with ChatGPT. Another with Claude. Another with Gemini. Another with a specialist not yet in the group. The swirl is the method; the work decides who leads.