Timeline

Built a native macOS menubar app in Swift for monitoring LaunchAgents. Used modern async/await concurrency, file watching, and SwiftUI. A foray into native development with Claude Code—proved it could work far beyond web apps.

Started building a knowledge management system for the Postcapitalist Design Club using Ollama for local AI processing. The key insight: sensitive community data should never leave the local machine.

Built an elaborate system with named pipes, fswatch, and 5 parallel augmentation agents. It was too complex. Learned: parallelism without clear coordination creates more problems than it solves. Simple text queues beat clever orchestration.

Built a website for Les Capitalistes Anonymes with radical simplicity and low carbon footprint as design goals. Delegated code cleanup and optimization tasks to Ollama—local AI works well for ancillary work like formatting, minification, and housekeeping.

Discovered Claude Code could control macOS apps via AppleScript, not just write web apps. Built Safari scraping utilities. Later added Keynote automation (key rule: use placeholders, not text boxes), Calendar access without OAuth, and Mail.app integration.

Built 12+ specialized Claude Code personas for different life domains: Simon (knowledge), Alan (operations), Brad (job search), Stuart (bookkeeping), Danny (education). Each worked well alone, but scaling across all of them created emergent complexity. No protocol for coordination. State fragmented across folders.

Built an agent to run the Postcapitalist Design Club. Started as a filesystem-based CRM using symlinks to create human-friendly folder names against UUID-based data folders. Later migrated to Obsidian to make it more human-accessible. Alan and Simon remain in daily use today—Alan running operations, Simon tending the philosophy.

The Postcapitalist Design Club workshops produced 20+ project ideas: cowifi (neighborhood fiber sharing), miam/popotes (community cooking), cuicui-fm (neighborhood radio), hola-vote (voting tools), marché commun (commons marketplace), stopminute (rapid response), bartr (barter network), musical chairs (economic games), and many more. Not part of Blue Box itself, but their ethos directly informed its philosophy.

Started building an agent to manage bookkeeping for a French consulting company (SAS). The vision: coordinated specialist agents for invoice generation, expense recording, income reconciliation. Explored Beancount for plain-text accounting. Remains unfinished—too ambitious for the time available, but demonstrated the potential for multi-agent coordination in complex administrative domains.

Built an agent to study my writing across published articles, replies, and drafts. Not to generate content, but to understand how I write—sentence structure, vocabulary, rhetorical devices, voice variations across contexts. An introspective research project: using AI to hold up a mirror, not to replace the writer but to understand him better.

Researched 35+ venues across Paris for POCA meetings—with complex constraints: free or cheap, evening availability, 12-24 people, RER access, aligned values. Discovered Paris Anim' centers (FREE for nonprofits), MVACs, La Bellevilloise (Paris's first cooperative, 1877), and solidarity economy hubs. A model for how Claude Code can assist real-world campaigns with multi-dimensional constraint sets.

Cataloged 66 themes and 540+ slide layouts in Keynote's AppleScript dictionary. Discovered the key rule: always use placeholders, never create arbitrary text boxes. Templates exist for a reason—they encode design intelligence. Fighting them creates chaos. Working with them produces consistent, professional results with minimal code.

Built run_websearch.sh: an intelligent web search using Ollama and ddgr that avoids CAPTCHAs, performs multi-round searches, extracts content, and provides AI-powered summaries. Can handle complex research queries requiring synthesis across multiple sources. A self-hosted alternative to cloud-based search tools.

After multiple failed orchestration attempts with local AI, discovered a key methodology: test ONE assumption per experiment. The previous failures—parallel agents with named pipes, chained AI calls, complex coordination—all tried to prove too much at once. Simple text queues, sequential processing, focused experiments. What works is boring. Cleverness is the enemy of completion.

Extracted and organized the Cooper typeface files for the Blue Box website—an open-source revival of Oswald Cooper's 1920s design by Indestructible Type. Converting between formats, understanding web font loading, choosing warmth over clinical precision.

A Postcapitalist Design Club workshop project: a discourse analysis tool that detects rhetorical manipulation, cognitive biases, and strategic omissions. Key insight that flows into Blue Box: tools should help people think, not think for them. "Treating people as reasoning agents, not targets."

Started articulating what Blue Box should be. Wrote the FOUNDATION.md document capturing core ideas.

To stress-test the vision, we convened an imaginary council of thinkers: Ivan Illich (on tools and conviviality), Simone Weil (on attention and grace), Audre Lorde (on self-care and resistance), E.F. Schumacher (on appropriate technology), Ursula K. Le Guin (on anarchist coordination), James C. Scott (on seeing like a state), bell hooks (on love as practice), Adrienne Maree Brown (on emergent strategy), Timnit Gebru (on ethical AI), Yuk Hui (on cosmotechnics), and Aaron Swartz (on information liberation).

Their critiques shaped the foundation. Their questions remain open.

Created this website to document the Blue Box project publicly. Initial content covers Philosophy, Vision, Nature, Experience, and Governance layers.

Deployed to bluebox.pocadesign.org. Fixed various path and permission issues.

Separated the conceptual documentation from the project log. Added stub pages for future layers: Specification, Architecture, Technology.

Switched from monospace to Cooper typeface. Warmer colors, better typography.

Started reconstructing the full timeline of Blue Box development by mining old Claude Code transcripts and session history. The goal: tell the real story of how this project evolved.