← agent-aaron.ai

programme operating model.

This page describes how the agent-aaron.ai programme is organised: the projects it runs, the layers they sit in, and the connective tissue between them. Explore the components of the system below, then read how the programme runs across its repeatable flows. The operating model is itself one of the programme's outputs, on the condition that the programme emits work others can use, and the programme is an independently built instance of the wider personal-AI-assistant-building movement it also studies.

The system.

How the programme runs.

The components above are static structure. The work happens in a small set of named, repeatable flows that run across them. Each flow below shows its stages and where each stage runs.

starter promptappendstarter promptsessionopensstart skillsession worksession-close skillDecisions LogTask TrackerSession LogFailure Modesnextsession
FLOW-4 · Session protocol

Closed loop: session-close emits the starter prompt that the next start consumes — continuity across threads is structural, not a memory exercise.

Session protocol

Every session runs inside a closed loop enforced by two skills, so no session begins without orientation and none ends without its state captured. The loop closes by design: Close prepares what the next Start carries forward, so continuity across threads is a structural guarantee rather than a memory exercise.

StartEstablishes what the session is for, maps that intent to the right model and effort, scopes the Notion reads the session needs, and emits a structured starter prompt. It sets the session up and does no work itself.
WorkThe session executes against that starter prompt, reading and writing Notion state as work happens.
CloseLogs every significant decision to the Decisions Log, updates every task in the Task Tracker, writes a session log entry, and produces the starter prompt for the next thread.

Design principles.

One home per projectEvery project lives in Chat or Cowork, never both, determined by the nature of its work.
Shared services, not mirrorsCowork projects serve the whole programme and are subordinate to no single Chat project.
Notion as the integration layerAll cross-project data flows through Notion; no project talks directly to another.
Skills enforce consistencyRepeatable processes are codified, with the specification as the source of truth.
Creation is separate from surfacingResearch and analysis projects create; the Website presents.
Lean over complexTools and structure are added only when they earn their place.
Cowork is additive, not load-bearingIf Cowork capabilities prove limiting, the Chat projects continue as-is, with no single point of failure.
A research testbedThe architecture is deliberately also primary source data for Redefining Change. The programme is itself a participant in the personal-AI-assistant-building movement, an independently built instance of the patterns that movement is converging on, so it both studies the movement and instantiates it.
Apparatus is subordinate to outputThe operating model is itself a legitimate output, on the condition that the programme periodically emits transferable work. Effort runs on two co-equal, permanent tracks: field-facing currency and transferable output. The separating test: does a change alter what the programme knows or can show about the world, or only how tidily it runs?

Current state & what's planned.

Established and in daily use
The governance model, the Notion integration layer and the core databases.
Research, operating with its substrate and synthesis modes in place.
The Intelligence pipeline, fully wired and running: union fetchers, a shared central store on Drive, and the Product Radar and Applied generators as scheduled Cowork tasks.
ProfileManager's three-layer artefact model, and Redefining Change working through discovery and early-tier analysis.
The website, live on Netlify, with the homepage, About section, Research and Redefining Change pages, and this operating-model page published.
The website copy architecture: the Site Pages database and per-project content briefs.
Planned, called out rather than implied as done
Extending the pipeline with an email fetcher as the next retrieval leg, then philosophical and further synthesis modes.
Migrating the website's files to a GitHub repository with automatic deployment.
A dedicated apps page for capability-demonstrator builds, gated on the first app going live, with proving runs of the app build pipeline alongside it.
The Website's remaining surfaces: an evolution timeline, and curated research and redefining-change content.
Documenting the website copy flow in the Flow Register, and growing the Studies and Themes substrate.
A foreground shift toward producing transferable output once the operating model and its public surfaces reach a standard Aaron is ready to stand behind.