Thematic analysis of the garden: fresh reading
Seven themes and one emerging pattern discovered by loading all 244 garden files into a single context window with no prior structure.
Part of experiment-2-thematic-analysis-with-1m-context.
Method: All 244 content files (articles, field notes, seeds, experiments, videos, weblinks, library entries, tags) loaded as raw markdown into a single 1M context window. No embeddings, no key phrases, no tag structure consulted. Themes coded purely from reading the text.
Theme 1: the quest for truth in a post-truth interface
The garden’s most persistent thread. Starts with Frankfurt’s definition of bullshit applied to ChatGPT, runs through Grice’s cooperative principle, surfaces in the DALL-E debiasing series (where prompt rewrites introduce new biases while claiming to remove old ones), and culminates in the designing-for-doubt concept. The concern isn’t that AI lies — it’s that it doesn’t care about truth, and that conversational interfaces exploit our Gricean trust. The “knowledge as missing fundamental” timpani metaphor connects here: truth as something you can approximate but never pin down in a probabilistic system.
Evidence: why-chatgpt-is-bullshit-and-why-we-should-design-for-that, bye-bye-alpaca-knowledge-as-the-missing-fundamental, air-canadas-bot-mishap-pre-dates-chatgpt, an-evening-with-youchat-and-chatsonic (the fake hyperlinks moment), de-biassing-dall-e, fingers-crossed-is-a-figure-of-speech-dall-e, llm-hallucinations-knowledge-as-missing-fundamental, designing-for-doubt, stay-calm-and-keep-thinking-for-yourself
Theme 2: the voice of the maker
A deep concern with authorship, authenticity, and what it means to write in an age of generation. The garden itself is a statement: “100% Maai” as a label, the AI transparency system (four levels), the insistence on writing as craft. The library tells this story too — Pratchett is the gold standard precisely because of voice: humor, layers, philosophy, language. The seed about the disappearance of authentic voice online names this fear directly. Nick Cave’s critique of AI-generated songwriting resonates. The article about writing with AI documents the friction of accordion editing and apple picking — trying to write through an AI rather than with it.
Evidence: youre-not-married-to-your-texts, ai-as-writing-partner, the-disappearance-of-authentic-voice-online, ai-transparency-in-content, what-happens-when-ai-has-read-everything, conversational-interfaces-are-not-easy, a-digital-garden-as-central-space
Theme 3: community as infrastructure
Maaike’s professional life is built on communities, not products. Convoclub (700 members), Voicelunch Language & Linguistics, Women in Conversational, Clubhouse communities — each described not as marketing but as the work itself. The freelance journey article makes this explicit: building communities has become the work. Even remote onboarding is described as community-building. The Convoclub insights article is original research: 300 answers analyzed, five challenges mapped.
Evidence: 2-years-and-700-members-insights-from-convoclub, convoclub-clubhouse, the-long-and-winding-road, the-lone-convo-designer, nieuwe-editie-voicelunch
Theme 4: the garden as epistemology
The garden is not just a website — it’s a way of knowing. The maturity system (draft/developing/solid/complete), the typed relations (laid paths, root systems, wind-seeded connections), the stable map positions, the distinction between tags-as-topic and relations-as-structure — all of these are epistemological decisions. The three connection layers (wiki-links = intentional, develops = structural, embeddings = emergent) mirror three ways of knowing: deliberate reference, structural belonging, and unexpected resonance.
Evidence: typed-relations-as-garden-infrastructure, stable-map-positions, digital-gardens-vs-blogs, thinking-in-public, principles-behind-this-garden, knowledge-gardens-and-serendipity, thematic-analysis-as-interaction-model-and-research-method, a-digital-garden-as-central-space
Theme 5: design is not engineering
A recurring insistence that what looks like technical work is actually design work. Context engineering? Let’s call it design. Prompt design? It’s information architecture, instructional design, topic-based writing — things that have existed for decades. The conversation about whether designers should learn programming holds both sides but keeps returning to: did I really help my customers with this? The Blender article argues that ethics should be designed into bot personas, not filtered on top.
Evidence: context-engineering-lets-call-it-design, putting-the-design-in-prompt-design, convo-question-should-i-learn-programming, 7-new-skills-for-conversation-designers-2022, blender-turned-rogue, scrum-for-conversational-teams, how-do-i-become-a-conversation-designer
Theme 6: the bilingual practitioner
Dutch is not incidental — it’s a lens. The DALL-E articles reveal how English-default AI produces cultural stereotypes when prompted with “Dutch.” The Dutch quantization seed shows how model compression silently damages non-English languages. Several articles exist in both Dutch and English. The library contains Dutch fiction and non-fiction alongside English. SSML tutorials are in Dutch. This isn’t just bilingualism but a critique position: being non-English-default in an English-default AI world reveals things that English speakers can’t see.
Evidence: de-biassing-dall-e, dutch-aware-quantization, hoe-word-ik-conversation-designer, onthaast-je-voice-action-met-ssml, SSML quickstart videos (Dutch), library books in Dutch
Theme 7: the knowledge graph as personal research program
What started as smarter backlinks has become the garden’s most technically ambitious thread. Apple’s Saga papers, embedding model surveys, key phrase extraction, similarity thresholds, spectral graph theory, blocking strategies — this is a research program. The observation box in the threshold tuning article captures the moment of realization: discovering a living textbook based on the garden’s own goals. The research gap seed goes further: nobody has studied embeddings for PKM formally. The garden could be both experiment and documentation.
Evidence: knowledge-graph-for-the-garden, reading-notes-saga-knowledge-graph, embedding-models-for-the-garden, tuning-the-similarity-threshold, content-chunk-size-and-embeddings, key-phrase-extraction-for-the-garden, full-scale-key-phrase-extraction, explore-page-design, stable-map-positions, typed-relations-as-garden-infrastructure, embeddings-for-knowledge-gardens-research-gap
Emerging theme: the non-linear mind
Scattered throughout but never addressed head-on: Maaike’s self-description as a non-linear thinker. The presentation prep article (rambling into ChatGPT voice mode as a creative process), the “troepjes” folder, the corporate escape narrative (“my managers would remark ‘you just seem to hop from topic to topic’”), the garden itself as a medium that fits a non-linear head, the book recommender’s attempt to solve “picking the next book is surprisingly hard.” This might be the garden’s origin story at a deeper level than the digital garden concept: the garden exists because linear formats don’t work for this mind.
Evidence: chatgpt-presentation-prep, the-long-and-winding-road, a-digital-garden-as-central-space, book-recommender