What is already working
- Native SwiftUI interface.
- GRDB-backed SQLite persistence.
- Calendar-driven home screen and Kanban views.
- Freeform canvas with connections.
- Recurring tasks and Reminders.app integration.
- A sibling-app data link to Aerial View.
Working product · case study
card&board
A native macOS planning application I conceived, specified, designed, tested, and now use every day—implemented in SwiftUI with Claude Code.
Working product, not a concept
The app on my Dock replaced the subscription tool I used to pay for
card&board is a native macOS task and planning app built around my own categories, card logic, planning horizons, and decisions about what should be destructive, reversible, local, or connected to the system.
It is not publicly released yet, but it is finished enough to be operational: I use it every day for active projects, deadlines, notes, recurring work, spatial planning, and the work that now appears across this site.
Designed before prompting
The mockup was the specification
Before the first prompt went to Claude, I had designed the interface in Penpot, screen by screen: the calendar, four levels of card collapse, priority colors, Kanban columns, archive flow, tag system, detail panels, and destructive states.
Every implementation round was reviewed against that specification. When a project bar became a pill, a priority rendered in the wrong color, or a sidebar collapsed to the wrong width, it was treated as a defect and corrected.
AI wrote the production code. I defined the product logic, interaction model, constraints, and acceptance standard before implementation began—and remained the design authority throughout the build.
That is the key difference between prompting toward an attractive result and directing a product from intent through implementation.
The product in use
Native interaction, spatial planning, and detailed task control
Human in the Middle
The human sets the intent, shape, and standard before the AI writes anything
The AI did the engineering. I stayed responsible for the product thinking, design authority, and judgment about what was actually right: why the app has no login or cloud sync, where deletion should be destructive, how the sibling-app relationship should work, and why Kanban moved off the canvas during development.
Human in the Middle does not mean supervising at the end. It means remaining the active reviewer of every meaningful decision between specification and working product.
The development record
- About a week and a half of active build sessions.
- Thirteen commits at the time of this case study.
- Sixty-one dated sections of decisions, corrections, and rebuilds.
- Repeated screenshot review against the Penpot specification.
- A working app used in real daily conditions.
Augmented product designer, not vibe coder
AI compressed the translation layer. It did not replace product judgment.
01 — Concept first
The problem, workflow, information architecture, screens, behaviors, constraints, and visual system existed before implementation.
02 — Authority throughout
Implementation was reviewed against a specification, and architectural decisions remained human-owned rather than being accepted as model defaults.
03 — Proof in use
The result is not a generated demo. It is a persistent native application that replaced a paid tool in everyday work.
What changed
The gap between “I know what this should be” and “this now exists” became thin enough to cross
Before large language models, I could specify this product in exhaustive detail but could not independently turn it into a native SwiftUI application with persistent data, drag and drop, calendar logic, a freeform canvas, and system integrations.
AI made that implementation reachable. The leverage came from combining it with an already formed product discipline: the ability to define the problem, design the whole system, recognize when the implementation is wrong, and keep responsibility visible.
Not publicly released · already in daily use
Designed deliberately. Built with AI. Owned and used as a real product.
card&board is a sibling app to Aerial View, a non-destructive file and asset manager built through the same process. A budgeting and invoicing app designed to work with card&board is now in development.