Journal

• Systems

  AI may scale judgment, but constraints still move

  Music, movies, and software scaled by building once and repeating. AI may scale judgment — but reliability, energy, and physical limits just move the ceiling. The central challenge is not eliminating problems. It is managing where the constraints move next.

  October 5, 2026 · 7 min read

• Economics

  The newest AI model isn’t always the best

  The newest, most accurate model can still cost more per finished task. What matters in production is cost per output that actually passes. Everyone compares models on price per token, or on which one is newest. In production, the better question is what it costs to get one output that actually passes.

  October 5, 2026 · 6 min read

• Alignment

  AI can drift away from the original problem

  AI can write, plan, and execute, but longer tasks lose the original intent. Checkpoints, not one perfect prompt, keep the work on the problem. The output can still sound confident and look polished while solving a different problem. Working with AI is less like giving a command and more like managing a project.

  October 5, 2026 · 7 min read

• Prompting

  AI is shaped by probability, not human judgment

  AI is not random, but it is shaped by probability. Prompt order, context, and tools can bias the answer — so important work needs review. Depending on how you prompt it, you can bias the response. Thinking models help, but they do not fully solve drift. For important work, restating the goal and checking the output is the process.

  October 5, 2026 · 8 min read

• AI Safety

  Why AI agents need limited access and clear objectives

  Unrestricted AI access and vague objectives create risk. Limit systems agents can touch, define the task clearly, and keep human judgment on important decisions. The Terminator story is not about AI becoming evil. It is about an objective followed too literally, with too much system access and no human judgment in the loop.

  July 20, 2026 · 6 min read

• Brand & Agents

  AI agents need personas that match the product arena

  Users expect agents to match the brand and interface they are in. Inconsistent tone, tools, or policy application force people to adapt to the AI instead of the product. McDonald's, Tim Hortons, Apple, and fine dining each set different expectations. AI agents break trust the same way a mismatched waiter breaks a restaurant — when persona and environment disagree.

  July 15, 2026 · 6 min read

• Expertise

  Where expertise sits when AI makes knowledge cheap

  When AI makes base knowledge cheap, value shifts to judgment, context, edge cases, and creative application. Expertise is where “what still matters” begins. Base knowledge is easier to retrieve than ever. The scarce skill is knowing which truths apply, where they break, and what to do when the context is messy.

  July 10, 2026 · 7 min read

• Human Advantage

  AI is practiced mindspace, not superior intelligence

  Experts and models both improve through repetition over a knowledge base. What people still own is judgment, behavior under pressure, and consistent execution in context. It is not that AI is "more intelligent" in every sense. It has been trained and forced to reiterate a vast knowledge base. Mindspace is not the same as judgment.

  June 28, 2026 · 6 min read

• Verification

  Why AI agents need human verification before they act

  AI can generate fast, but speed without oversight creates liability. A human-in-the-loop layer turns raw AI output into something you can actually trust. The question is no longer whether AI can do the work. It can. The question is whether you are comfortable letting it act on your behalf without anyone checking first. For most teams, the honest answer is no.

  March 10, 2026 · 7 min read

• Workflow

  Building expert review workflows for AI outputs

  Routing AI-generated work to the right reviewer is not a manual process anymore. Structured review workflows turn ad-hoc checking into a repeatable system. Most teams already review AI outputs informally. Someone eyeballs a draft, edits it, and sends it. That works until volume increases and the person checking is no longer the right expert for every type of output.

  February 14, 2026 · 6 min read

• Governance

  How audit trails create trust in AI-powered operations

  When every review action is recorded, trust stops being a feeling and starts being a verifiable fact. Audit trails are the foundation of accountable AI. Accountability is not about slowing AI down. It is about making it possible to look back at any decision and understand who reviewed it, what they decided, and why. That record is what separates experimentation from production.

  January 20, 2026 · 5 min read

• Use Cases

  Why everyday AI users deserve expert verification too

  AI is no longer just for developers and enterprises. Regular people rely on AI for resumes, contracts, and financial decisions, and they deserve the same verification layer. A parent drafting an IEP dispute letter, a freelancer generating a contract, a first-time homebuyer summarizing mortgage terms. These are real decisions with real consequences, and most people have no way to know if the AI output they are trusting is actually right.

  December 5, 2025 · 8 min read