Hi, I’m Björn.
AI Content Lead by day, vibe coder by curiosity, and the creator of ALTRUIXM. This is my small corner of the web.
I design the workflows, prompts and quality gates that turn AI from a novelty into a dependable part of content production.
I build apps by describing what I want and iterating with AI until it works — shipping real things without a traditional developer background.
ALTRUIXM is my own desktop app for analysing content, finding gaps and writing from them — built entirely by vibe coding, and used daily on my own work.
My project
A macOS desktop app that analyses existing content, finds the gaps, and writes new content from those findings — with a deliberate human checkpoint in between. AI prepares the decision; a person makes it.
Two tracks run through this story: what the app became, and what I learned building it without a formal development background.
March 2026
The app
Development began on a Claude Pro license with two separate prototypes. One analysed existing content: it scraped, embedded, clustered, and found gaps. The other wrote new content from those findings.
What I learned
You can get surprisingly far by describing outcomes instead of writing code. You also build up a lot of undocumented decisions very quickly.
Spring 2026
The app
The two prototypes became a single macOS desktop app with two processes and a deliberate human checkpoint between analysis and writing.
What I learned
The checkpoint became the core design principle. AI should prepare the decision, and a person should make it.
Alongside
The app
Experiments in video generation and in orchestrating several tools into one content pipeline.
What I learned
Roughly half the bugs only appeared in the packaged app, never in development mode. Since then, every build phase gets tested both ways.
July 2026
The app
The project moved under git, and a run of releases followed: a Mac-native interface, a model picker, per-project style guides, and support for skill files.
What I learned
Version control is not bureaucracy. It is how a solo builder remembers why things changed.
July 2026
The app
A release stopped AI-generated drafts from feeding back into the source corpus.
What I learned
If AI writes the whole content chain over time, output converges on earlier AI output. Voice rules fix tone but not this drift in substance. I wrote about it on LinkedIn.
August 2026
The app
The earlier prototypes were retired. ALTRUIXM became one standalone product with versioned releases, used daily on my own work.
What I learned
Using your own tool every day is the most honest test there is.
Late summer 2026
The app
Gap detection was redesigned, not just retuned.
What I learned
Similarity measures what a text is about, not what it actually says. Checking whether something is already covered is a different job from suggesting new angles, so the two need different methods.
October 2026
The app
I gave Claude Code more autonomy over development. The app moved to local-first storage and gained its own local tool interface (MCP).
What I learned
The bottleneck had shifted from writing code to deciding. Delegating the build loop freed me to focus on product direction.