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 3 of 8
Clear filtersAIWhich AI Agent Actions Need Human Approval FirstNot every agent action needs a human, and gating all of them kills the point. Here is a checklist for which agent actions to approve and which to let run.OperatingTell Customers Before You Change the ModelSwapping the model under your AI product silently is a governance failure. Why model changes are breaking changes and how to notify customers the right way.ThesisYour Books Must Be Explainable, Not Just AutomatedAutomated books are worthless if you cannot explain them. Why explainable bookkeeping beats speed, and what it takes to defend every number to anyone who asks.EngineeringWhy AI Agents Need Immutable LogsAI agents need immutable logs because an editable record is not evidence. Append-only, tamper-evident logging is what makes an agent's actions provable.OperatingHow to Set Quorum Thresholds for Governance VotesSet a governance quorum too high and nothing passes; too low and a handful decide everything. Here is how to size quorum thresholds that make votes legitimate.ThesisThe Myth That AI Agents Go RogueThe fear that AI agents go rogue is mostly Hollywood. The real risk is boring: agents do exactly what you told them, on bad instructions, with too much access.ThesisGoverning AI Agents vs Models: Autonomy Raises the BarGoverning an AI agent is harder than governing a model. When a system acts, not just answers, your governance needs new controls. Here is what changes and why.StrategyLiquid Democracy: Vote Delegation That WorksLiquid democracy lets voters delegate their vote to someone they trust and take it back anytime. Here is how vote delegation fixes low turnout without central control.ThesisWhy Regulated Industries Are the Best AI MarketRegulated industries are the best AI market because governance is a moat there. Where trust is mandatory, capability alone loses and disciplined vendors win.OperatingThe First 24 Hours After an AI Incident: What to DiscloseAI incident disclosure in the first 24 hours decides whether customers trust you again. Here is what to say, when to say it, and what silence actually costs.StrategyHow to Publish Honest AI Benchmarks Buyers TrustCherry-picked AI benchmarks fool no serious buyer. How to publish honest benchmarks with real methodology, failure cases, and numbers that survive scrutiny.EngineeringHow to Handle Disputed Prediction Market OutcomesSooner or later a market resolves and someone contests it. Here is how to design a dispute process for prediction markets that settles fairly and keeps trust.