Writing/2026-05-19
Why modeless exists as a test ground for small, strange, useful AI products that do not fit cleanly into existing categories.
Most interesting AI products will not start as venture-scale companies.
Some will be too small. Some will be too weird. Some will be useful to one person, one team, one industry, one recurring document, one public dataset, or one operational bottleneck that nobody outside the room has ever heard of.
That does not make them unimportant.
AI changes the economics of this kind of software. It lowers the cost of building, maintaining, adapting, and explaining tools that would previously have been too narrow to justify. A product can now live in the long tail: specific, useful, inspectable, and small enough to stay close to the problem.
modeless exists for that territory.
The first wave of AI products has been easy to name: chatbots, copilots, search tools, writing assistants, coding agents, meeting summarizers, image generators.
Those categories matter, but they are not the whole story.
The more interesting layer is the long tail of AI-native software: tiny tools, private utilities, domain-specific copilots, agent-facing services, document processors, workflow surfaces, public-data demos, generated interfaces, and strange hybrids that do not yet have a clean product category.
Some of these things may become commercial products. Some may stay as experiments. Some may only exist to clarify a pattern.
That is fine. The point is to make them visible enough to learn from.
We do not yet know the final forms of human-AI interaction.
Sometimes the right interface is a chat. Sometimes it is a dashboard. Sometimes it is a generated UI, an approval gate, a trust receipt, a spreadsheet that turns into software, a background monitor, a document review surface, or an API that an agent can call without a human ever looking at it.
These are different software modalities: different ways software can appear, behave, and participate in work.
That phrase sounds abstract, but the practical question is simple: what shape should software take when the user is not only clicking buttons, but delegating, supervising, correcting, approving, and asking systems to reason across messy context?
modeless is a place to try those shapes.
The site is organized as a small ecosystem because the work does not all belong in one bucket.
Products are the more durable artifacts: named tools, services, and systems that may keep maturing.
Experiments are allowed to stay unstable. They can be fragments, demos, prototypes, or visualizations that are useful before they are finished.
Writing gives the work a memory. It captures decisions, build notes, cost curves, product lessons, and the ideas that are easier to understand in prose than in a screenshot.
The design system keeps the whole thing from becoming a pile of unrelated sketches. It gives the artifacts a shared language: sharp surfaces, visible state, readable metadata, trust cues, and enough structure to make AI-native interfaces feel inspectable rather than magical.
modeless is intentionally product-led. The work should show up before the biography does.
That is the point of the catalog, the experiments, the design system, and the notes. They are not just content sections. They are a way to test what kinds of AI products want to exist when the cost of building software keeps falling and the boundaries between documents, workflows, agents, and interfaces keep moving.
Some ideas will be practical. Some will be non-commercial. Some will be public infrastructure. Some will be small enough that they only need to work once for the right person.
That is the long tail I want modeless to make room for.