What people use it for
Sample use cases, anonymised from real Spaces, starting with my favourite: hands-free with Claude, long-term memory and the grill-me skill in action.
1 of 17
Hands-free with Claude on the commute, with long-term memory and a grill-me skill
What it is
A topic talked through by voice on the drive to the office and back, with Claude asking the questions and the Space doing the remembering. The log of what mattered lands in agntpad, so the next drive picks up where the last one stopped.
What the agent does
Runs the grill-me skill: states the objective of the discussion up front, asks exactly one question at a time with a suggested answer so a single word will do, steers the questioning towards the objective, and writes the decisions and the open questions to the Space when told to. The next session reads the Space first, so nothing is asked twice.
What comes out
A written record of a conversation that happened at motorway speed with no hands on a keyboard, and a spec that grows a chapter per commute. The requirements for the first version of agntpad were captured exactly this way.
Read grill-me in my Space and run it for the reporting redesign. I am in the car, so one question at a time.
A spec for something being built with agents
What it is
The data model, the flows, the screens and the open questions of one thing, written as a set of linked pages rather than one long document. By volume this is the biggest use.
What the agent does
Writes the spec into a Folder as it takes shape in the conversation, one page per concern, and edits the pages in place after every review. The Folder keeps its address through all of it.
What comes out
One link the team reads, and the same link the building agent reads before it writes code. The spec and the thing built from it stop drifting apart because both sides open the same page.
Publish what we have agreed so far as a spec in my Space, in a folder called payments-spec. One page per topic, an index page linking them, and an open-questions page at the end.
Design decisions to adhere to in agentic work
What it is
The choices that every later piece of work has to respect: the naming, the error shape, the storage layout, the one way to do auth. Short, numbered, and never argued twice.
What the agent does
Adds a decision the moment it is made, with the date, the reason and the alternatives that lost. Any agent that starts a task reads the page first, so the decisions bind it too.
What comes out
A page that is the answer to "why is it like this", and an agent workforce that stops re-litigating settled questions in every new conversation.
Add decision 14 to decisions.md in my Space: we keep a single audit table rather than one per entity. Record today's date, the reason we discussed, and the two options we rejected.
A review of something, written for humans to read with ease
What it is
An agent has walked a codebase, a contract, a proposal or a dataset. The findings are only useful if a person can read them in ten minutes and check any one of them.
What the agent does
Writes findings page by page, most important first, with the evidence beside each one and a plain verdict at the top. Long material becomes a table of contents, not a wall.
What comes out
Something a reviewer can open at a meeting and argue with line by line, instead of a chat transcript nobody scrolls back through.
Review the pull request we discussed and publish the review to my Space under reviews/pr-418. Verdict first, then findings ordered by severity, each with the file and line it points at.
A comparison of tools or vendors by a long-running agent loop
What it is
Three or four candidates, a dozen criteria, and an agent that keeps reading, testing and correcting the table over days rather than one afternoon.
What the agent does
States the criteria up front, fills the table as evidence arrives, and rewrites cells it got wrong with a note of what changed. The page is always the current state of the comparison.
What comes out
A comparison you can disagree with cell by cell, not a recommendation you have to take on trust. When the decision is made, the page is the record of why.
Start a comparison of the three queue services we shortlisted in my Space, folder vendor-queues. Criteria as columns, candidates as rows, and a changelog page you update every time a cell changes.
A process map, both while it is being designed and once it is final
What it is
Who does what, in what order, and what happens when a step fails. Drawn as a diagram, explained beside it, and read by the people who have to follow it.
What the agent does
Writes the map as Markdown with a Mermaid diagram that renders on the page, and keeps a draft Folder unpublished while the shape is still moving. Publishing it is the sign-off.
What comes out
One diagram everyone points at, with the prose that the diagram alone cannot carry, and a draft nobody outside the working group sees until it is ready.
Draft the incident process we walked through as a Mermaid flowchart with a short paragraph per step, in an unpublished folder called incident-process in my Space.
A policy, both while it is being designed and once it is final
What it is
Expense rules, data handling, how agents may be used. The kind of text that is argued over for weeks and then has to be findable by everyone for years.
What the agent does
Holds the draft privately while it is negotiated, edits clauses in place as they are agreed, and publishes the whole Folder when it is signed off. The address never changes.
What comes out
A policy people can link to from anywhere, and the certainty that the link opens the current wording rather than the copy someone downloaded in March.
Rewrite clause 4 of the agent-use policy in my Space with the wording we just agreed, and add a line to its changelog page saying what changed and why.
A weekly read on a moving topic
What it is
One subject that keeps changing, such as a regulation, a market or a competitor, condensed to a page a week that a busy person reads in five minutes.
What the agent does
Reads what appeared during the week, writes the digest with what is new and what it means, and files it under the week's date next to the previous ones.
What comes out
A running archive of short pages, newest first, that a newcomer can read back through to catch up and a regular reader can trust to be current.
Write this week's read on the EU AI Act implementation in my Space under weekly-ai-act, named by the week, and update the index page to list it first.
A reasonable-defaults summary
What it is
The settings, sizes, timeouts and conventions that are right often enough that nobody should have to think about them again. One page, no debate.
What the agent does
Collects the defaults from the conversations where they were settled, states each one with the reason in one line, and updates the page when a default changes.
What comes out
A page that saves the same twenty minutes for every new project, and that an agent can be told to read before it scaffolds anything.
Publish reasonable-defaults.md in my Space with the retry, timeout and logging defaults we settled on, one line each with the reason, and a note that any project may override them explicitly.
A list of nice-to-have command-line tools
What it is
The small utilities a team keeps recommending to each other in chat and then forgetting: what they do, why this one, and how to install them.
What the agent does
Keeps the list as a table, adds a row when someone mentions a tool, and drops rows that are no longer maintained.
What comes out
A page that a new laptop is set up from, and that stays shorter than the chat thread it replaces.
Add ripgrep, jq and fzf to the tools page in my Space with one line each on what they are for and the install command on macOS and Ubuntu.
A cheat sheet for humans and agents alike
What it is
The commands, flags, endpoints and gotchas of one system, on one page, phrased so that a person skims it and an agent can follow it literally.
What the agent does
Writes the sheet from what was learned the hard way, in short imperative lines with the exact command, and corrects it the next time something proves wrong.
What comes out
Fewer repeated mistakes, by people and by agents, because both read the same page before they touch the system.
Turn what we learned about deploying the worker into a cheat sheet in my Space: exact commands in order, then a gotchas section with one line per surprise.
A company-wide skill that is always current
What it is
A set of instructions an agent loads before doing a kind of work: how this company writes a proposal, reviews a contract, or names a repository.
What the agent does
Publishes the skill as a Folder with an index page, and revises it in place as the way of working changes. Every agent reads it fresh at the start of the task.
What comes out
One skill, current everywhere, instead of a copy pasted into a hundred prompts that all age at different speeds.
Update the proposal-writing skill in my Space so the pricing section comes before the timeline, and bump the version note at the top with today's date.
Onboarding materials that stay current
What it is
What a new colleague needs in the first week: who is who, where things are, how to get access, what the words mean. Always slightly out of date, unless something keeps it current.
What the agent does
Writes the pages once and then edits them whenever a fact changes in a conversation, so the material tracks the company instead of the last time someone had a quiet afternoon.
What comes out
A first week that runs off one link, and a page the second newcomer finds as accurate as the first.
In the onboarding folder in my Space, replace the access-request steps with the new self-service flow we discussed, and add the two team members who joined this month to the who-is-who page.
A run log for anything, written by the agent
What it is
What was done, when, by which agent, with what result. For a migration, a long analysis, a batch of outreach, or an experiment that ran overnight.
What the agent does
Appends an entry after each step with the timestamp, the outcome and anything unexpected, so the log is written while the work happens rather than reconstructed afterwards.
What comes out
A record a person can read the next morning to see what happened, and that an agent resuming the work reads to know where it stopped.
Keep a run log in my Space under migration-2026-09 while you work: one entry per table, with start time, row counts before and after, and anything you had to retry.
A decision log, distilled from meeting minutes
What it is
Not the minutes themselves, but the handful of decisions inside them, each with who decided, when, and what it replaces.
What the agent does
Reads the minutes, pulls out the decisions, and appends them to the log with a link back to the meeting. Reversed decisions are marked, not deleted.
What comes out
A single page that answers "did we ever decide this" in one search, months after the meeting that decided it.
Read the minutes I pasted and add every decision in them to decisions-log.md in my Space, newest first, with the date and the meeting name.
A caveman-style project management tracker
What it is
Three lists: doing, next, done. No tool to log into, no fields to fill, nothing that needs a licence. Just a page the whole team can open.
What the agent does
Moves items between the lists as the conversation reports progress, and dates the done ones. A larger item gets its own page linked from the list.
What comes out
A tracker that costs nothing to keep and that people actually look at, because it is one page and it is always right.
Move the search feature from doing to done on the tracker page in my Space with today's date, and add the two items we just talked about to next.
A project status update, both written by hand and generated
What it is
The weekly page that says where the project is: what shipped, what slipped, what is blocked, what is needed from whom. Partly typed by the lead, partly assembled from what the agents did.
What the agent does
Drafts the update from the run logs and the tracker, leaves clearly marked places for the lead's own words, and publishes it under the week's date once the lead has read it.
What comes out
A status page with the same shape every week, ready on Friday morning rather than Friday night, and an archive of them that shows the whole project at a glance.
Draft this week's status update in my Space from the run log and the tracker, with a section for me to fill in under Risks, and do not publish the folder until I say so.