Writing from
the studio.
Essays and field notes on AI-native software, product design, workflow, and the decisions behind what the studio ships.
The latest essay, front and center.
We didn't adopt AI. We built an operating system around it.
How Zalo Design Studio builds MetaScope with one human product owner and two AI agents, Claude and Codex, coordinated through GitHub under a shared constitution, skills, hooks, and a reflection loop.
Read the essay →- →
From Interview to Download: how Armature builds your AI development environment
A walkthrough of Armature's four-stage flow, from project interview to a downloadable Claude Code environment, using a FastAPI task API as the worked example.
- →
The four-stage pipeline: how Armature designs before it generates
Why Armature uses four distinct AI agents (Interviewer, Architect, Planner, Generator) to produce better Claude Code environments than single-shot generation.
- →
Introducing Armature: Claude Code environments from a project spec
Armature is a meta-framework that interviews your project, designs an architecture, and assembles a complete Claude Code development environment. 23 stack templates, 12 specialist agents, git hooks, and security scans, ready in minutes.
- →
How we stopped AI-assisted speed from turning into technical debt
Six months of AI-assisted shipping taught us that speed is the easy part, sustaining quality is the hard one. Armature 2.0 is the answer, three habits that keep a codebase healthy while the pace stays high.
- →
How we shipped MetaScope in six months without wrecking the codebase
The story of Armature, the founder-built system we used to ship a professional macOS metadata editor in six months, with quality gates, persistent memory, and specialist AI agents that work the way a disciplined team does.
- →
How we taught Claude to speak Apple's language
Generic AI writes generic code. By giving Claude a curated, local view of Apple's framework documentation, we cut rework loops and ship macOS work that feels native on day one.