Desktop agent - define the user experience first; the per-token API is an acceptable starting path and model-integration choices stay open

Situation

At the 10/26 Desktop Brainstorm Greg pushed on definitions and billing (a desktop harness on the API means per-token cost and no Codex or Claude Code subscription use). Peter set the working rule: settle what the agent experience should be first, treat the API as a known-workable path, and do not block on how CIQ interfaces with the model - more options (local harness APIs, routing layers) will open before launch. The experience he described: bring your own agent or key, with granular OS permissions (full disk access yes or no, specific folders) granted up front. He opened by restating the 10/21 decision that stock GNOME is a zero-investment placeholder to be retired and work is not doubled.

Reasoning

The integration landscape is changing month to month, so locking a plumbing decision now locks it on the worst information. CIQ value is the OS-side contract - hooks, permissions, what is exposed and protected - which holds regardless of which agent the user brings. Bjorn owns the product vision; Peter owns the engineering sequencing of what to decide now and what to leave open.

Additional Context

Confirmed by Peter. The GNOME placeholder statement is a reinforcement of 10/21 and is folded in here rather than logged separately.

Observed Evidence

Step one is figuring out what we want it to do. / I want us defining the hooks, the APIs by which the agent can interface with the operating system. / do I give the agent full disk access? Yes or no. If I say no, do I give it access to these folders.

Matching Patterns

Confidence Breakdown

Reasoning Depth Analysis