Dossier 03 · AI Systems & Orchestration

AI as work infrastructure.

The durable value of AI emerges in systems that organise knowledge, prepare decisions, produce work and verify their own results.

System architecture

Models are orchestrated by function, not brand preference.

Research, architecture, implementation, criticism and verification receive separate responsibilities.

Facts, decisions and handovers live in versioned artefacts rather than ephemeral chat history.

Writing steps, costs and irreversible decisions receive clear approval and stop points.

Results count when supported by data, logs, tests or read-back verification.

Knowledge systems

Sources are captured structurally, placed in time and marked by evidence level.

Conflicts, gaps and duplicates are treated separately rather than silently smoothed over.

Decisions, lessons and handovers remain available across sessions, models and devices.

Privacy, secrets, rollback, access and accountability are built into the architecture.

Working principles

A system’s success message is a signal. Evidence decides.

Criticism and solution remain separate so that no instance can approve its own work.

The architecture should still carry when individual models, tools or agents are replaced within months.

Describe the situation. The right form of collaboration follows from the problem, not from a standard package.