Writing · Nº 041065 essays
The writing
Ideas from inside the work. Explore technology, entrepreneurship, and the practice of building things that last.
93 essays found · Page 6 of 8
Clear filtersEngineeringWhat to Log for an AI Feature, Field by FieldExactly what to log for an AI feature: inputs, model version, prompt, output, and decisions. The field-by-field record that makes an AI answerable later.StrategyWho Resolves a Prediction Market, and Can You Trust ThemEvery prediction market needs someone to declare the outcome. Here is how market resolution works, the oracle problem, and how to pick a resolver you can trust.ThesisTrust Is the One Thing You Cannot RebuildAcross a portfolio carrying one name, a single confident lie does not break one feature. It breaks the reason anyone trusted the whole system. That asymmetry sets the rules.OperatingWho Should Own AI Governance in Your CompanyAI governance fails when everyone owns it, which means no one does. Here is how to assign real ownership across engineering, legal, and the person accountable.EngineeringKeep a Research Agent From Citing Sources That Don't ExistA research agent that invents citations is worse than no agent. Here is how to force source-grounding so every claim traces to a real, checkable passage.EngineeringHow to Make Every AI Decision Defensible After the FactHow to make AI decisions defensible: capture inputs, versions, and reasoning at decision time so you can prove what happened months later when it matters.ThesisPrompt Injection Is a Governance Risk, Not Just a BugPrompt injection is a governance risk you must disclose and control, not a bug you patch once. How to handle it and prove you handle it to buyers.StrategyQuadratic Voting: When It Actually Makes SenseQuadratic voting lets people express how much they care, not just what they prefer. Here is when it beats one-person-one-vote and when it quietly fails.OperatingHow to Run a Union Ratification Vote OnlineA union ratification vote needs a secret ballot and a verifiable count at the same time. Those are not in conflict. Here is how to run one that holds up.ThesisWhy a SOC 2 Report Does Not Prove Your AI Is SafeSOC 2 proves you have controls, not that your AI behaves. Here is the trust myth that gets AI vendors in trouble and what buyers should actually ask for.ThesisWhat to Ask an AI Vendor About Their Audit TrailThe audit trail questions that separate real AI vendors from demos. Ask these before you sign, because you cannot bolt an audit trail on later.ThesisHow to Red-Team Your AI Product Before Buyers DoRed-team your AI product before enterprise buyers do it for you. A practical method for finding jailbreaks, leaks, and failure modes on your own terms.