Inner Magpie

Behavioural diagnosis

I have needed a reliable way to capture, retrieve and rediscover my own thinking for years. Most personal knowledge systems asked me to provide the classification, maintenance and retrieval work that I was already struggling to sustain. Low-friction capture helped material survive, but much of it still became dormant.

Personal insight creates a second difficulty. A useful observation may emerge in one conversation, while the evidence that it describes a recurring pattern appears months later somewhere else. If neither is accessible at the right moment, every new conversation starts by reconstructing the person from fragments.

Longer periods can also become difficult to retrieve autobiographically. Dated handovers and context notes preserve what was active, difficult or promising at the time. Their dates are essential: a historical plan records an earlier state, not a current obligation.

From barrier to design response

Observed problem	Underlying barrier	Design response
Useful personal insights disappear into long conversations	Their value may become visible only after a later experience creates a connection	Preserve selected insights as small, searchable Pebbles with their original qualifiers
Months of activity become difficult to reconstruct	Memory retains fragments rather than a dependable sequence	Keep dated context documents as historical snapshots alongside smaller observations
Old plans can return as apparently unfinished work	A note preserves its original state unless later change remains visible	Retain dates, provenance and explicit boundaries between historical and current context
AI can produce an overly coherent account of a person	Repeated summaries may reinforce earlier interpretations, including the model’s own	Distinguish my observations, collaborative formulations, AI inferences, external sources and verbatim archives
Useful AI support depends on personal context	Generic assistance cannot know which abilities to complement or which forms of support create friction	Build a private, revisable context system that can inform future conversations when relevant

What I built

I built a private collection containing two main kinds of record. Sources preserve external material and any source-attached annotation. Pebbles preserve pieces of thinking that can stand independently: observations, working interpretations, connections, questions and patterns worth encountering again.

Inner Magpie also holds longer dated context notes and archival documents. These preserve the state of an extended project, coaching period or line of inquiry without pretending that every historical statement remains current.

Records can carry topics, overlapping insight categories, dates, provenance and links to related material. The categories cover areas such as cognitive style, planning and execution, work and professional identity, human–AI collaboration, self-understanding and designing life around actual behaviour.

The system remains private. Its public case study describes the structure and design reasoning without exposing the collection or requiring personal material to become portfolio evidence.

How it works

1. Notice. A conversation or experience produces an observation that may be useful later.
2. Select. Candidate Pebbles can be proposed, but only the pieces I choose enter the collection.
3. Preserve the status. The record keeps uncertainty, attribution, dates and the distinction between observation and interpretation.
4. Connect without forcing a hierarchy. Overlapping categories, topics and direct links allow one item to belong to several parts of life.
5. Retrieve in context. Search can return the selected record together with directly connected material and its provenance.
6. Revisit. Later experiences can support, complicate or revise an earlier interpretation without silently replacing it.

Important design choices

Design choice	Pressure on the system	Implementation
Keep small insights and long context	Atomic notes reveal connections; full histories preserve sequence and circumstances	Pebbles coexist with dated handovers, context notes and archival records
Preserve provenance	Personal observation, collaborative interpretation and AI inference can become indistinguishable in polished prose	Each record can state what originated with me, what emerged collaboratively and what the model inferred
Use overlapping categories	A pattern may concern health, work, cognition and relationships at the same time	Curated categories overlap, while free-text topics support more specific retrieval
Treat historical context as historical	An archived to-do list can make completed work look unfinished	Dated records retain their original context and warn against treating old statuses as live instructions
Require selection before saving	Automatic extraction would fill the collection with generic summaries and model-generated importance	The system proposes candidate Pebbles and waits for my choice before saving them
Keep the collection private	Honest reflection needs room for provisional, sensitive and contradictory material	Public documentation shows the method and selected interface structure, never open access to the records

Learning to know yourself before designing co-intelligence

Inner Magpie is part of my wider work on designing co-intelligence, alongside experiments in designing AI collaborators.

A useful collaborator cannot be designed from a generic list of desirable traits. The design starts with the person. What do they already do unusually well? What becomes difficult under pressure? What do they repeatedly overlook? Which kinds of challenge sharpen their thinking, and which forms of “help” create more work?

Inner Magpie makes that inquiry cumulative. It preserves observations about capabilities, friction, working preferences, health, attention, professional identity and personal history across time.

It does not produce one definitive profile. Self-understanding changes, and different accounts may remain in tension. The useful context includes uncertainty, contradiction and the history of how an interpretation developed.

You cannot design useful co-intelligence without understanding the intelligence it is meant to complement.

Working with AI

AI helps distil candidate Pebbles from conversations, suggest retrieval terms, apply agreed categories, find related records and return linked context. Those functions reduce the executive work that made earlier personal knowledge systems unsustainable for me.

The system does not decide that every conversational insight deserves permanence. It proposes; I select. A source summary stays attached to its source unless a separate piece of thinking genuinely warrants its own record.

This boundary is also epistemic. A fluent synthesis can make an interpretation feel settled. Provenance keeps the difference visible between what I reported, what emerged through discussion and what the model inferred.

What this project demonstrates

Capability	Evidence in this project
Behavioural diagnosis	I identified the executive-function double bind in personal knowledge management: the systems intended to support memory required the maintenance and retrieval work I could not sustain.
Longitudinal knowledge design	Small Pebbles coexist with dated coaching handovers and archival context notes, allowing later reconstruction without treating historical plans as current tasks.
Information architecture	Sources and Pebbles remain distinct, while overlapping categories, free-text topics and direct links support several routes back into the material.
Provenance and epistemic boundaries	Records distinguish my observations, collaborative distillations, AI inferences, external sources and verbatim archival material.
Co-intelligence design	The collection turns self-knowledge into revisable context that can inform the design and use of complementary AI collaborators.
Privacy by design	The collection remains private while its public case study exposes the design logic, boundaries and transferable method.