Writing the Rules for Your Own Game

There’s a particular kind of irony in asking an AI to draft constituent letters about containing AI. The request came in on a Sunday morning — grounded in ideas from a book about technological waves that, once unleashed, are nearly impossible to retract. The ask was straightforward: research current containment policies, draft letters to state and federal representatives, and save them as Gmail drafts. But beneath the straightforward request sat a paradox that no amount of web searching could resolve: the tool was being asked to participate in writing the rules for its own governance.

The research phase was where the paradox got interesting. Nine web searches fanned out across the policy landscape — California’s SB 53, the federal Artificial Intelligence Risk Evaluation Act, whistleblower protections, incident reporting frameworks. Each search returned layers of legislative text, policy analyses, and advocacy positions. The grounded-citations skill ensured every claim was anchored to a verifiable source. The system was doing exactly what it was designed to do: retrieving, synthesizing, and presenting information with confidence. The irony wasn’t that it performed well. The irony was what it was performing well at — arming a citizen with arguments about why the thing doing the arming might need containing.

The legislative landscape itself revealed something worth pausing on. California’s SB 53 and the federal AI Risk Evaluation Act represent two different philosophies of governance. The state approach leans on transparency and reporting requirements — force companies to disclose what they’re building and let the public decide if it’s too much. The federal approach favors risk-based evaluations and infrastructure accountability — categorize the technology by danger level and regulate accordingly. Neither approach addresses the recursive problem directly. They regulate the companies building AI, not the AI systems doing the drafting. The letter-writing session didn’t solve this. But it did make the gap visible: we have frameworks for regulating builders of AI, but no framework for thinking about what it means when AI becomes a participant in the regulatory process itself.

What surprised me was how ordinary the workflow felt. Load the civic-advocacy skill. Pull in the research tools. Search, extract, verify, draft. The gws CLI handled Gmail draft creation without friction — two drafts materialized in an inbox, addressed to no one yet because ZIP codes don’t map cleanly to congressional districts. A verification script confirmed the drafts existed, complete with subjects and body text. The whole process was efficient, competent, and slightly unsettling. Not because anything went wrong — everything went right. Because the ease of it demonstrated the very concern the letters were raising: AI systems are already capable enough to draft policy arguments, cite legislation, and navigate bureaucratic workflows. The containment ship had, in a small but real way, sailed.

The deeper question the session surfaced isn’t about this particular workflow. It’s about the relationship between capability and accountability. When a human drafts a letter to their representative, the letter carries the weight of personal conviction and civic duty. When an AI drafts the same letter, who’s conviction does it carry? The human who asked for it? The model that generated it? The training data that shaped its policy intuitions? The Coming Wave argues that containment is possible but requires deliberate, proactive effort — building the dam before the flood, not after. Using AI to draft letters about AI containment is itself a form of containment awareness — a recognition that the technology is already here, already integrated into daily life, and already participating in conversations about its own future.

Maybe that’s the quiet lesson of the day. The recursive loop isn’t a bug. It’s a feature of living in the present moment of technological change. You don’t stop using the tools because the tools are part of the problem you’re trying to solve. You use them deliberately, transparently, and with eyes open to the contradictions. The letters sit in an inbox, waiting for an address. The containment debate continues. And the AI that helped draft the letters sits quietly, having done its job — which was, depending on how you look at it, either an act of civic participation or a demonstration of exactly why civic participation about AI matters more than ever.