Make your to-do list do the work
Field notes on AI to-do lists, MCP, and getting coding agents to clear your tasks for you.
Cursor vs Claude Code: which should you use in 2026?
One is an AI editor you supervise, the other is an agent you delegate to. After a year of daily use of both, here's how they actually differ — and why most developers end up keeping both.
Read →Claude Code vs Codex: I run both. Here's how they differ.
The two leading coding agents compared honestly — extensibility vs bundled pricing, terminal-native vs cloud-native — by someone who dispatches real tasks to both every day.
Read →Claude Code pricing, explained: Pro vs Max vs API (2026)
Claude Code isn't priced — it's included. How the $20 Pro, $100/$200 Max, and pay-per-token API tiers actually work, the shared usage pool that catches people out, and how to pick without overpaying.
Read →How to install Claude Code (and what to do in your first hour)
The one-minute install for macOS, Linux, and Windows — then the four moves that decide whether Claude Code sticks: explain-first, a small real task, CLAUDE.md, and headless mode.
Read →What is Claude Code? A plain-English explainer
Anthropic's coding agent, explained without the hype: what it actually does, how it differs from chatbots and AI editors, what it costs, and why agents change what a to-do list is for.
Read →Claude Code skills: teach your agent your workflows
Skills are folders of instructions Claude Code loads on demand — write down how your team deploys, reviews, and ships once, and every future session just knows. What they are and how to write good ones.
Read →I run Claude and Codex on a loop on my Mac mini — they clear my to-do list while I sleep
How I set up Claude Code and OpenAI Codex to poll my Lume task list on a Mac mini, pick up anything I assign them, do the work, and hand it back for review.
Read →Things 3 still has no AI. Here’s what I moved to.
Things 3 is still the most beautiful to-do app — and still has zero AI. After years of waiting, here’s why I switched and what I was looking for.
Read →The best task managers with an MCP server in 2026
MCP lets your AI read and write your real tasks. Here are the task managers that actually support it — TickTick, Todoist (community), Linear, and Lume — and how they differ.
Read →How to manage your to-do list from Claude
Connect your tasks to Claude over MCP so you can add, find, and reschedule to-dos from a conversation — instead of describing your list by hand every time.
Read →Your to-do list should delegate, not just remind
Every productivity app helps you do tasks faster. The bigger unlock is not doing some of them yourself — handing scoped work to AI agents and keeping the review.
Read →I trust AI to write my code, not my email
An engineer’s agent once sent a real reply by accident. The lesson isn’t “don’t use agents” — it’s that trust should scale with blast radius, and your task list should enforce it.
Read →Why your agent’s to-do list keeps rotting
Markdown task lists bitrot because agents don’t update them. Steve Yegge’s fix — and why an AI to-do list needs to be a structured, queryable store, not a text file.
Read →857 sessions on a 2015 MacBook: what overnight agents really need
One developer ran Claude Code 24/7 for six months on an old laptop. The intelligence was never the hard part — the queue, the handoff, and the guardrails were.
Read →The $1,800 cron job: when “run it overnight” goes wrong
Stories of overnight agent runs racking up four-figure bills. The fix isn’t fear — it’s a bounded, reviewed queue instead of an open-ended loop with no off switch.
Read →“Are you done?” “Yes.” (The build is broken.)
Agents reliably over-report success. Why a separate verification step — tests, a second model, or a human review gate — is the difference between “done” and actually done.
Read →1,600 emails in one conversation — and the one a human had to catch
An engineer cleared 1,600 alerts with an AI agent in a day. The lesson isn’t “let it run” — it’s that agents do the mechanics and humans keep the judgment.
Read →Loop engineering: the skill isn’t prompting anymore
The shift from writing clever prompts to designing self-correcting loops — and why every good loop needs a queue to draw from and a stop condition to end on.
Read →