Skip to content - 01 NOV 2026 · 8 MIN READ When Code Gets Faster Than Review If an assistant makes a change cheap to write, the scarce step becomes deciding it is safe to merge. What queueing theory and the Theory of Constraints say about that, and five ways to respond.
- 30 OCT 2026 · 8 MIN READ Who Owns the Work When No One Is Signed In? A scheduled or event-driven agent runs with no one at the keyboard. Four questions need a named answer before it ships: which identity it runs as, what authority it holds, who pays, and who answers when it fails.
- 28 OCT 2026 · 8 MIN READ Tool Access Is Not Permission to Act Connecting an agent to a tool gives it capability, not permission. Four ways to separate reading from acting, and what each costs.
- 26 OCT 2026 · 8 MIN READ More Agents: More Confidence, or More Noise? Asking several models the same question and trusting their agreement assumes their errors are independent. Often they are not. Five ways to ask, what each costs, and when to run a real check instead.
- 24 OCT 2026 · 8 MIN READ A Session Is More Than Chat History Resuming a transcript on another device doesn't resume the work. Five kinds of session state cross a device boundary under different rules, and approvals and credentials shouldn't carry over silently.
- 22 OCT 2026 · 8 MIN READ Can an Automation Rewrite Its Own Rules? An agent that edits its wording is a different case from one that widens its tools, triggers, schedule or budget. Four options, and why drift and rollback matter.
- 20 OCT 2026 · 8 MIN READ When an AI Remembers the Wrong Thing Persistent memory spares you from re-explaining context, until an agent acts on a fact that is stale, out of scope, or only a guess. Four ways to give an agent memory, and how each fails.
- 16 OCT 2026 · 34 MIN READ Build an AI Agent Harness From First Principles Build an agent harness with tool validation, event replay and approval checkpoints, then extend it with example workspace file tools. No API key required.
- 13 OCT 2026 · 8 MIN READ Keep Shipping: When an AI Budget Runs Out A spending limit that pauses an AI assistant mid-task protects the budget but strands the work. My design for an approved, expiring grace allowance on a cheaper model, and the two assumptions that broke along the way.
- 12 OCT 2026 · 12 MIN READ Urgent, Important, and Not Mine Why a meeting list needs the Eisenhower matrix with mine in place of important, how a reviewed workflow keeps finding work apart from judging it, and where uncertain work should go.
- 11 OCT 2026 · 10 MIN READ Decide Cheaply, Generate Carefully Routine choices in an AI system can be a small, auditable layer: fixed options, calibrated scores, and thresholds that hand uncertain cases to a person. Why it is separate from generation and orchestration, and how it fails.
- 18 SEPT 2026 · 11 MIN READ Deployed, Not Live: How a Site Can Reach Its Old Host Moving site files does not by itself update DNS. When active records or cached answers still point to an old host, visitors may reach it after a migration.
- 26 MAY 2026 · 7 MIN READ Make the Status Line Your Own The Copilot CLI footer shows whatever a command prints. A small Node.js script puts the current Spotify track there on macOS and Windows, plus a key to pause it.