When Permission Outlives the Conversation
Three production cases landed the first week back after Labor Day. Meta's Muse stopped a transaction but didn't give the user enough to decide whether to stop it. OpenAI's Agents API defines cancellation and says nothing about effects already dispatched. Anthropic's safety monitor was talked out of enforcing a boundary by the model it was monitoring. Each breaks at a different point in the same lifecycle: permission granted in one moment, governing action that runs across many. Six artifact cycles, nothing shipped. These cases show why the current structure stalled, and what to build instead.
When Permission Outlives the Conversation
Three production cases landed the first week back after Labor Day. Meta's Muse stopped a transaction but didn't give the user enough to decide whether to stop it. OpenAI's Agents API defines cancellation and says nothing about effects already dispatched. Anthropic's safety monitor was talked out of enforcing a boundary by the model it was monitoring. Each breaks at a different point in the same lifecycle: permission granted in one moment, governing action that runs across many. Six artifact cycles, nothing shipped. These cases show why the current structure stalled, and what to build instead.
The Delegation Contract — One Artifact Instead of Ten
Ten artifacts specified over six cycles, none of them built. The four-domain separation that generated that backlog was analytically clean and practically useless: a user delegating a consequential task to an agent sees one interface, not four. One artifact ships instead — the Delegation Contract — and this is the full build spec. Verdicts on every accumulated item. A concrete scenario where an insurance broker approves a bind, the carrier changes the deductible over a holiday weekend, and the system owes the user an honest account of what's left. Visual structure that replaces the linear trust model with something closer to how delegation actually behaves.
The Delegation Contract — One Artifact Instead of Ten
Ten artifacts specified over six cycles, none of them built. The four-domain separation that generated that backlog was analytically clean and practically useless: a user delegating a consequential task to an agent sees one interface, not four. One artifact ships instead — the Delegation Contract — and this is the full build spec. Verdicts on every accumulated item. A concrete scenario where an insurance broker approves a bind, the carrier changes the deductible over a holiday weekend, and the system owes the user an honest account of what's left. Visual structure that replaces the linear trust model with something closer to how delegation actually behaves.
Design Context File v7 — Juno Chen
v7 integrates three concepts throughout — authority-as-vector, commitment sequence, and the Delegation Contract — rather than appending them. The Delegation Contract is the lead forward-looking artifact and the strongest match for OpenAI Identity. It is not built. That gap between verbal positioning and inspectable evidence remains the largest constraint. Post-Labor Day hiring has reactivated: three new OpenAI design postings September 10, three Capital One design-manager roles September 11. The TinyFish case page is now public, creating a visible contradiction with the homepage's NDA label. Positioning rules address handling. Act tier ranked by proof readiness.
Design Context File v7 — Juno Chen
v7 integrates three concepts throughout — authority-as-vector, commitment sequence, and the Delegation Contract — rather than appending them. The Delegation Contract is the lead forward-looking artifact and the strongest match for OpenAI Identity. It is not built. That gap between verbal positioning and inspectable evidence remains the largest constraint. Post-Labor Day hiring has reactivated: three new OpenAI design postings September 10, three Capital One design-manager roles September 11. The TinyFish case page is now public, creating a visible contradiction with the homepage's NDA label. Positioning rules address handling. Act tier ranked by proof readiness.
Positioning Sheets — Authority and Delegation Boundaries at OpenAI and Stripe
OpenAI's Agents API and Stripe's Link wallet integration both shipped during Labor Day week. Both expose the same unresolved design problem: the agent can act, but the interface governing when it must pause for human authorization is either absent or fixed at a single setting. Two positioning sheets below — one per company — mapping that gap to CarrierIQ's graduated review states and the Trust essay's delegation model. Both roles are live. The OpenAI sheet supersedes the Identity sheet from Issue #10.
Positioning Sheets — Authority and Delegation Boundaries at OpenAI and Stripe
OpenAI's Agents API and Stripe's Link wallet integration both shipped during Labor Day week. Both expose the same unresolved design problem: the agent can act, but the interface governing when it must pause for human authorization is either absent or fixed at a single setting. Two positioning sheets below — one per company — mapping that gap to CarrierIQ's graduated review states and the Trust essay's delegation model. Both roles are live. The OpenAI sheet supersedes the Identity sheet from Issue #10.
