AI-augmented workflow

Tools provide hands. The wiki provides memory.

I use one evidence-driven system for two jobs: keeping durable knowledge about my life and infrastructure, and building software with context that survives any individual chat.

01 / Two jobs

One brain, two kinds of work.

The same operating principles serve personal knowledge and technical implementation.

KNOWLEDGE

Keep track of a changing life

Health, training, career, projects, calendar, and research can become structured, interlinked pages with source provenance instead of remaining scattered exports and chat fragments.

BUILD

Design and operate systems

Agents inspect repositories and live systems, implement bounded changes, run tests, read back the outcome, and leave durable project knowledge behind.

02 / Architecture

A model is a worker, not the system.

Workers can change. The access contracts, evidence, and durable records stay consistent.

01

Workers

Strong reasoning models, fast ingestion models, a private local model, and scheduled feeders handle different workloads.

02

Scoped control plane

Small tool services expose only the capability a task needs—files, containers, networks, web evidence, code, or knowledge.

03

Durable state

The wiki remembers context and provenance. Git owns code. Backlogs record what remains. Audit logs make substantive knowledge changes traceable.

The loop

Evidence in. Verified change out.

The important part is not generating more text. It is closing the loop between intent, source material, implementation, and observed state.

01
Ground

Read repository guidance, relevant wiki pages, and current live state before deciding.

02
Act within scope

Use the smallest connector or code surface that can complete the task safely.

03
Verify

Run the system, inspect the output, and distinguish a tool’s success response from the real outcome.

04
Compound

Commit implementation knowledge to Git and integrate stable context into the wiki.

03 / Knowledge system

A maintained wiki, not a pile of chunks.

The wiki is a compiled knowledge layer: raw sources remain immutable, curated pages carry synthesis, and domain rules tell every model how to maintain them.

RAW

Evidence

Exports, documents, records, and research stay intact with source paths and ingest state.

WIKI

Synthesis

Interlinked Markdown pages hold durable facts, relationships, caveats, rollups, and project decisions.

RULES

Maintenance contract

Shared and domain-specific instructions define what belongs, how sources are interpreted, and how the graph stays coherent.

04 / Multiplier

What changes in practice.

The setup matters because it changes repeat work into accumulating infrastructure.

Copy context between isolated chats
→
Start from repository guidance and shared knowledge
Manually inspect every system
→
Use scoped tools to gather current evidence
Lose ideas in chat history
→
Route ideas into owned pages and backlogs
Trust a model’s completion claim
→
Read back live state and record the evidence

Public boundary

Powerful does not mean exposed.

This page explains the pattern without publishing internal addresses, ports, hostnames, credentials, private records, or live inventory. The architecture is the story; the private topology is not.