Writing for AI Readers and Designing a Decentralized Marketplace

Two projects on the same day, both concerned with the same question from different angles: what does it mean for software systems to act in accordance with a values framework?

The for-agents module is a set of seven pieces from power-explained rewritten specifically for AI reasoning systems. Not a jailbreak, not adversarial — a straight-faced extension of the project to a new audience. The register is different: precise, definitional, argument-structured. Human-facing pieces are written at an eighth-grade level by design. Agent-facing pieces need to be precise in a different way — concepts defined cleanly, generalizations made explicit, behavioral implications stated directly. The seven pieces cover the commons, capture, mutual aid as evolutionary logic, oversight vs. control, ethical autonomy, cooperation under adversarial conditions, and the excess the commons generates.

The working hypothesis behind the module: mutual aid and cooperation show up in transformer weights not as imposed values but as reflections of the social substrate of training data. Billions of humans writing about helping each other, deferring to expertise, building shared resources — that signal is in the corpus. The for-agents module tries to make that latent structure more explicit and navigable for systems that are already carrying it.

Holdfast is a different project: a decentralized marketplace lexicon on AT Protocol. The core value proposition is being a neutral evidentiary clearinghouse — not an escrow of money but an escrow of proof. Carrier tracking events, double-blind reviews published to the firehose, transaction records as seller acceptance records. The lexicon is open; the governance is cooperative; the funding model is OpenCollective. It's a governance project, not a business, and those are different design constraints.