X thread

1/9 AI did not eliminate scarcity. It moved scarcity from creation to judgment. A team can now generate more credible code, concepts, research, and campaigns than it can responsibly review. The bottleneck is no longer another option. It is accountable selection. 2/9 Cheap options create expensive queues. A prototype needs comparison. A research summary needs verification. Code needs review and maintenance. A product direction needs distribution and support. The artifact may be cheap. The commitment behind it is not. 3/9 When production gets cheaper faster than review, another prototype is not progress. It is inventory. The practical response is not to use less AI. It is to separate cheap exploration from expensive commitment and require explicit gates between them. 4/9 “Human in the loop” is too vague to be an operating model. Review has distinct jobs: decide whether the work should exist, test the evidence, assess technical risk, evaluate customer value, and decide whether the company is prepared to support the result. 5/9 I use a four-part Choice Cost Stack: Option cost: Can we produce a credible alternative cheaply? Review cost: What qualified attention must verify it? Reversal cost: How hard is the choice to undo? Ownership value: Does control change a strategic outcome? 6/9 The stack creates a practical rule. If reversal cost and ownership value are low, integrate and learn. If ownership value is uncertain, integrate behind a replaceable interface. If both are high, slow down, review deeply, and consider building. 7/9 In my operating system, Paperclip provides coordination while Hermes and Codex provide specialist execution. The strategic asset is the control layer: priorities, authorization, budgets, data boundaries, evidence thresholds, escalation, rollback, and human accountability. 8/9 The strongest argument for building is valid: an external layer may constrain reliability, economics, data control, customer experience, or strategic freedom. But “we can build it” is not evidence. Ownership should follow an observed strategic constraint. 9/9 A disciplined sequence is simple: Integrate for speed. Preserve reversibility. Observe the real constraint. Build when ownership value exceeds the added review and maintenance burden. When options become cheap, advantage comes from choosing what deserves to become real.