ClarkCant · Engineering journal
Ideas, experiments and experiences.
Inside the making of ClarkCant. Read the thinking, play with the details, and see how it works.
- 01
Inside ClarkCant: how one message becomes an answer, a widget or a task
A source-level tour of ClarkCant: one runtime and many clients, a conductor that decides, Pi behind a single adapter, widgets as a vocabulary, a policy that only narrows, memory you can delete, and voice that is only a voice.
- 02
This blog runs on ClarkCant's ideas
Articles are structured bilingual documents, widgets are the same ones Clark draws in chat, every post has a Markdown twin for AI readers, and agents write through scoped, revocable MCP access. Here is how it fits together.
- 03
The art of minimalism: either I'm crazy, or this app is
AI was supposed to make software simpler. Instead we learn a new word every day. On the obsession behind ClarkCant, what I deliberately left out, and why I might be wrong.
- 04
Autonomy without losing control
An agent can act on its own while the human stays the final authority. How ClarkCant does it with policy, provenance, audit, reversible actions, an honest "I don't know", recovery and hard resource limits.
- 05
Why AI agents should stop asking for permission
If an agent asks before every command, every file and every API call, where exactly is the autonomy? Why ClarkCant treats your request as the authorization, and the few places it still stops and asks.
- 06
Build in public: what changed in ClarkCant over the last two weeks
An honest walk through the git history from 22 September to 5 October 2026: an inbox for what waits on you, settings back in the conversation, widgets Clark can press, and the parts that still are not proven.