I’m building a practice in AI experience design: making AI useful beyond developer tools, while keeping its state, scope, and consequences understandable to the people using it.
My work moves between design, cognition, technology, media, and everyday routines. A desktop utility, an agent workflow, a local archive, a controller, or a small interface can all become ways to make an ordinary experience lighter and more capable.
Enter the Live Lab → · 한국어 · 日本語 · 简体中文
I tend to begin with friction that is easy to overlook: a change that cannot be reviewed, a folder that has become unmanageable, a language barrier, or a repeated action that should take one button.
The recurring question: how can AI help people feel more capable without making human choice disappear?
- AI experience design for people without a development background
- Cognition, development, learning, attention, and the ways tools reshape behavior
- Review-before-write agents that keep human decisions visible
- Local-first systems for private, durable, and reversible work
- Playful interfaces that make serious tools less intimidating
- Cross-disciplinary experiments without a fixed medium or product category
Open lab notebook — hypotheses, tensions, and unfinished questions
- AI becomes more useful when its state, scope, and uncertainty stay visible.
- Non-developers do not need simplified intelligence; they need better interaction contracts.
- A playful interface can reduce fear without hiding operational seriousness.
- Local-first architecture is not only privacy infrastructure; it is an experience-design choice.
| Tension | What I am exploring |
|---|---|
| Automation ↔ agency | How much should happen automatically before control becomes invisible? |
| Intelligence ↔ legibility | Can a powerful system still explain what it is doing in human-scale terms? |
| Utility ↔ delight | Can a serious workflow feel playful without becoming distracting? |
| Breadth ↔ coherence | How can open-ended exploration still produce focused, useful artifacts? |
- What does an AI interface look like when conversation is only one material among many?
- How can tools support cognitive growth instead of replacing effort indiscriminately?
- Which everyday frustrations are small enough to overlook but meaningful enough to redesign?
Reusable skills for turning Hermes Agent into a practical collaborator with explicit scope and grounded procedures. |
A Windows media workspace designed for useful content operations without requiring a developer-facing pipeline. |
A review-before-write interface that makes agent changes inspectable, scoped, and reversible. |
Korean localization treated as access design, backed by automated coverage and type verification. |
Controller shortcuts that turn repeated creative and media actions into physical memory. |
Defensive review utilities for web and local projects that are explicitly authorized. |
More tools from the workshop
- Hermes ComfyUI Maker — image workflows through conversational agents
- Vaultwright — local-first Obsidian maintenance
- BB TxT — fast local text search
- Bumboo Build Inspector — structured review for rapidly built software
Python, TypeScript, JavaScript, Node.js, Electron, PowerShell, GitHub Actions, Obsidian, Hermes Agent, MCP, and ComfyUI. I choose them according to the experience and constraints of the work; the tool list is not the identity of the practice.
The panel below is generated from public repository data. It shows a small factual snapshot rather than a score, streak, or productivity claim.
- Local first. Keep personal workflow data close to the person who owns it.
- Review before write. Make changes visible before they become durable.
- Reversible by design. Prefer drafts, branches, backups, and clear exits.
- Open-ended research. Let the question choose the discipline and medium.
- Everyday delight. Serious technology can still feel playful and humane.
Still exploring, prototyping, and crossing boundaries. More finished work and notes live at bumboo.fun.