Journal
A design discussion on what actually makes a set of AI teammates a team — role prompts, routed workflows, or genuinely separate worker contexts — using Solvie's five public roles as the working example. No benchmark claimed; the comparison worth running is set out, not conducted.
A prototype approval screen shows the right interaction: an AI agent drafts, a human reviews the exact text, and only a click sends it. That is not the same as a proven send-boundary. Using our own Miyaha prototype as the example, this note separates what the screenshots demonstrate from what a real boundary needs — server-side enforcement, changed-draft reapproval, exact recipient/payload binding, single-use approval, and revoked-access checks — and sets out an acceptance-test checklist for telling the two apart in any AI agent.
Short archived note on why Solvie One shipped as a public waitlist rather than a live product.