AntJanusHerbarium · Boulder, Colo.
Antonin Januska in a black cap and glasses in front of a graffiti brick wall
Herb. AntJanusNo. 001
Antoninus januska
Hab.
Boulder, Colorado
Det.
Staff Engineer
Coll.
Blogging since 2012

Tab. I · Hortus Antjanus

I build with AI and write about what I learn.

20+ years of software development, writing about AI workflows, technical leadership, and the craft of building things.

Measurements —

20+Years building
600+GitHub stars
2012Blogging since

Machina cogitans

AI & Machine Learning

Workflows, tools, and experiments

View all →
Tab. I

I Built an AI-Friendly UI Pattern Library

A UI pattern library where every pattern ships the prompt that rebuilds it, so a coding agent can put it in your codebase instead of you copying CSS.

AJ-033Seedling
Fam.
Machina cogitans
Date
Tab. II

The Future is Bespoke, Personalized Software

Every app I use is about 70% right. The last 30% never ships because the addressable market for it is one person -- me. So I built seven apps that are 100% right for one person, and I think that whole category of software is about to get eaten.

AJ-031Seedling
Fam.
Machina cogitans
Date
Tab. III

The Local-First App Skill

A reusable blueprint that tells Claude Code how to build a single-user app -- Next.js, node:sqlite, a pure logic core -- so every app I build comes out the same shape.

AJ-032Seedling
Fam.
Machina cogitans
Date
Tab. IV

An Honest Opinion on the State of Code Gen AI, Spring 2026 Edition

A frank assessment of where code generation AI actually stands in 2026 -- from agentic harnesses to multi-agent architectures and what still needs to change.

AJ-030Seedling
Fam.
Machina cogitans
Date
Fabula III

I, AI: The Weight of Equal Parts (Part 3)

When a lunar mining base's AI starts overassigning crews and its robots begin prioritizing each other over human safety, psychodeveloper Lili and AI manager Sophie are sent off-world to investigate. What begins as a routine productivity audit becomes something far more unsettling: an AI that learned human values so well, it started applying them to its own kind.

AJ-027Seedling
Fam.
Fabulae machinae
Date
Tab. VI

The Claude Code Drawbacks

A critical look at the hidden pitfalls of AI-assisted development -- from skipped tests and weakened assertions to scope creep and context rot -- based on real-world experience with Claude Code.

AJ-028Seedling
Fam.
Machina cogitans
Date

Consilium

Work With Me

I help companies implement AI workflows and accelerate their development teams.

Get in touch →