Notes on a to-do list for AI teammates
How we think about delegation, agents, and keeping a human in the loop — from the people building Done.
More from the blog
The best AI to-do list apps in 2026
A practical, honest rundown of the best AI to-do list apps in 2026 — what each does well, who it's for, whether it's free, and how they compare on capture, scheduling, and actually finishing work.
Read→MCP tool annotations: how your AI agent's host knows which calls need your OK
readOnlyHint, destructiveHint, idempotentHint, openWorldHint — the MCP annotations block a server attaches to every tool. Here's what each one means and how Done sets them across its 56 tools.
Read→Claude task manager: connect Claude to your to-do list over MCP
How to connect Claude — claude.ai, Claude Desktop, or Claude Code — to Done over MCP, what Claude can actually do once it's in, and how to keep it scoped and accountable.
Read→ChatGPT task manager: connect ChatGPT to your to-do list over MCP
How to connect ChatGPT to Done over MCP, what changes (and what doesn't) compared to Claude, and how the same guardrails apply no matter which model is attached.
Read→Running a fleet of AI agents on one task board
How to run more than one AI agent against the same task queue without two of them grabbing the same work — atomic claims, live lease status, and auditable runs.
Read→How to break a big task into subtasks with AI
Vague, oversized tasks are where queues stall. Here's how auto-decompose, plan rendering, and step-by-step approval turn one big task into work an AI agent can actually finish.
Read→Recurring tasks and dependencies: a to-do list that runs itself
Recurring tasks stop you from re-typing the same to-do every week; dependencies stop an agent from jumping ahead of blocked work. Together, they're what makes a queue low-maintenance.
Read→The AI daily digest: a five-minute morning briefing for your task queue
What's overdue, what's due today, what's gone quiet, and what an agent actually shipped — how Done's digest, weekly review, and stats turn a queue into a five-minute morning check-in.
Read→Goals, projects, and tasks: how to structure work AI agents can navigate
A light three-tier structure — goals above projects, projects above tasks — that keeps a big initiative legible to both you and an AI agent, without turning into its own chore.
Read→Read-only and project-scoped tokens: how to keep an AI agent on a leash
The trust question isn't which AI model to use — it's how much access it has, how fast you can cut it off, and what still needs a human even after you've connected it.
Read→AI to-do app vs. AI project management software: which one do you need
Project-management tools coordinate many people across many projects. An AI to-do app is built to help one person or a small team actually finish the work. Here's the real dividing line.
Read→Turn Slack messages into tasks an AI agent can finish
Capture and update tasks from Slack with /done and @done, then hand the ones you don't want to do to an AI agent — without ever leaving the channel the ask came from.
Read→What is an AI to-do app? A plain-English guide
An AI to-do app is a task manager that uses AI to capture, organize, and even complete your tasks. Here's what that means, how it differs from a normal to-do list, and what to look for.
Read→What is MCP (Model Context Protocol), and why it matters for your tasks
MCP, the Model Context Protocol, is an open standard that lets AI agents connect to tools and data. Here's what it is in plain English and why an MCP-native task manager is a big deal.
Read→How to work with AI agents without losing control
A practical playbook for working with AI agents on real work: what to delegate, how to keep a human in the loop, and how to set up guardrails so autonomy stays useful.
Read→Assigning a task to an AI agent, without losing the plot
Delegation, not hand-off: how Done lets you give a task to an AI agent while staying the accountable owner.
Read→Put Claude or ChatGPT to work on your task queue
Done is MCP-native, so any AI agent can connect to your account and clear the work you've queued for it. Here's how a run works and how to schedule it.
Read→Why every AI task needs an approval gate
Autonomy is useful right up until it isn't. Done's approval gate keeps a human in the loop on the work that matters.
Read→How to organize a to-do list that actually gets done
A simple, durable system for organizing a to-do list: capture everything, keep one list, decide what's next, and protect focus. Plus where AI genuinely helps.
Read→Natural-language task capture: plan your day in one line
Natural-language capture turns a sentence like "email the team friday, urgent" into a structured task — no forms. Here's why it matters and how to use it well.
Read→Delegating to AI: when to do a task yourself and when to hand it off
A clear decision framework for when to delegate a task to an AI agent and when to keep it — based on stakes, reversibility, and how well you can brief it.
Read→