THE WORK RETURNS.
You finish a brief, then turn to the idea it unlocked. By the time you return, the facts have moved. Your curiosity should not have to pay the full cost of starting again. Let the document carry a method for becoming useful again.
THE LIVING DOCUMENTS MANIFESTO
The researcher. The writer. The builder. The idea you have not found time to follow.
Let your documents carry more of the routine, so you can carry more of the ambition.
18 ideas · One FOLD at a time · Start reading01 / The conviction
You finish a brief, then turn to the idea it unlocked. By the time you return, the facts have moved. Your curiosity should not have to pay the full cost of starting again. Let the document carry a method for becoming useful again.
The researcher gathers sources. The builder needs a plan. The writer needs the argument behind it. A working folder can keep those threads close: the document, its recurring jobs and its review history. Give different parts of your work a place to meet.
Keep my project status brief ready for the weekly review. Show what moved, what remains and where I need to pay attention.
Read status.csv. Refresh the declared table and chart in the nightly window. Leave my own writing untouched.
Six of eight tasks complete, up from four. Inspect the source and run record. Keep the update or revert the changed part.
You refine the sources, instructions and review criteria. Keep the useful method with the working folder. Carry that learning into the next run.
An adaptation of Orionfold’s FOLD method. This sample uses Base data gathering and redraw. Night Shift acts within standing permission; you review afterward. You improve the method between cycles.
Begin with a brief, a budget or a plan you already care about. A source-backed update or a chart that redraws can leave more room for the next question. Small responsibilities are how a larger ambition becomes practical.
A FAMILIAR PLACE TO BEGIN
Selected uses among US employees who use AI. Gallup, Q2 2026; reported July 20. Categories overlap.
Respondents reported their uses of AI at work. Categories overlap. Bars use a common 0–60% scale. Read Gallup’s report
WHY THIS NEEDS TO CHANGE
Someone still has to verify it, connect it to sources and return when the facts change. The next generation of tools has to carry more of that continuing work.
In Microsoft's survey, 60% said they trusted AI output enough not to check its accuracy.
A polished draft can pass the checking work to someone else. Flow should keep the source, exact change and review decision close enough to inspect together.
Microsoft People Science · 2026-02-10Self-reported attitudes, not observed error rates. The underlying sample was mid-to-senior business roles at organizations with 1,000+ employees, not a representative prosumer sample. Correlations do not establish causality.
349 technical workers reported a median 3× speed gain, compared with 1.4–2× gains in the value of their work. METR cautions that perceived gains can exceed actual gains.
Producing more is an incomplete measure of progress. Flow should leave a visible result, revision history and review trail so you can judge what was useful.
METR · 2026-05-11Self-reported counterfactual estimates from a selected sample, not measured productivity or a randomized trial. Different questions produce the range of value estimates. This is not evidence that current AI makes all workers slower.
Across about 400,000 interactive Claude Code sessions, people made roughly 70% of planning decisions while Claude made roughly 80% of execution decisions.
The brief, boundaries and meaning of done still need an owner. Flow should make that intent durable when a recurring job carries work forward.
Anthropic · 2026-06-16Observational, model-classified transcripts; not causal evidence or direct observation of real-world outcomes. Non-interactive/headless sessions were excluded. Expertise and success classifications have measurement limits. Do not present strict transcript verification rates as product failure rates.
A Living Document should carry its context forward, make changes inspectable and keep a routine useful on its next cycle. More output is only a beginning.
Research context and our interpretation. These studies did not evaluate Flow. The challenge headlines above are Orionfold’s interpretation; the linked titles are the original publications.
02 / The principles
The document is a Markdown file you own. Read it, edit it, keep it. Your work should remain useful when you change a model, pause a subscription or choose another tool. Make the file the foundation.
Choose the files, folders and public pages the job may use. State which parts may change. A model, a network request and a script have different permissions. Delegation becomes practical when the assignment is specific.
Your argument.
Your interpretation.
The words you want to keep.
Your document is ordinary Markdown. A standing job is permission to do the work you declared.
Daytime Agency waits for your approval of the exact diff. A standing Night Shift job acts within permission you already gave.
An update should leave a trail: what ran, what it read and what it changed. Keep sources, Evidence and Checks beside the result. The strongest answer invites you to inspect how it was made.
No meaningful change is a useful result. A failed source is an exception to explain. A fluent rewrite is not proof of progress. Judge continuing work by the attention it saves, not the amount of text it produces.
Automation can do the upkeep. You decide what to keep. Flow’s Morning Briefing brings changed parts together for review, with Keep and Revert. A trustworthy routine makes recovery part of the experience.
Review a part. Keep the result or restore that part. Later edits are protected from a blind rollback.
Keep your attention. An unchanged source does not need a fresh paragraph.
See the exception. An unreachable source is an explained problem, never invented evidence.
03 / The practice
Point a competitor brief at a few public sources. Let Night Shift notice a change and refresh declared parts. Return to the passage, the evidence and your own notes. Pick up the argument where the world moved it.
A budget or status report can bind its tables and charts to data. When the input changes, the view can follow. Keep the visual, its definition and the underlying numbers together, so the result is easier to question.
| Workstream | Done | Total | | Research | 8 | 10 | | Launch | 6 | 8 | | Editorial | 3 | 6 |
Workstream, Done, Total Research, 8, 10 Launch, 6, 8 Editorial, 3, 6 View: completion by workstream
Sources → Definition Definition → Table Definition → Chart Table + Chart → ReviewIllustrative data and readable definitions.
| Workstream | Done | Total |
|---|---|---|
| Research | 8 | 10 |
| Launch | 6 | 8 |
| Editorial | 3 | 6 |
Writing, search, pictures, tables, 34 chart types and 20 diagram types are the familiar foundation. Explore the four-part Flow tour.
A local model can turn gathered data into Overnight notes. Read them against the numbers and follow the question they raise. The routine prepares a starting point. What you make of it is where your experience, curiosity and imagination enter.
Choose a page you actually reopen. Define one useful job in the editor. Run it once, inspect the result, then give it a rhythm. A small routine that earns repeat use matters more than a sweeping promise.
The source update reports six of eight launch tasks complete. The bound table is refreshed.
Gather → redraw the declared view“Launch is now six of eight tasks complete, up from four. Two fewer tasks remain.”
Illustrative model note · Pro requiredRead the note against its source. Keep it, or restore this part. Your next question is yours to choose.
Marked change → Morning BriefingSample workflow. Deterministic gathering and redraw are Base; model-powered notes are Pro. Start with one job you understand.
04 / The standard
Give recurring work a place in the day. Flow’s background helper can run with the app quit, while your Mac is plugged in and idle. Choose a nightly window. Return to a Morning Briefing that helps you begin.
Better models can widen what you attempt. Clear assignments, useful sources and thoughtful review give that capability somewhere to go. Keep the method visible, so each improvement can strengthen a practice you understand.
A clear assignment can become a routine.
Flow can be open or quit. Your Mac still needs to be available.
Gather data, refresh declared parts and prepare local model notes when configured.
A changed document is ready to inspect. A quiet result can stay quiet.
A missed window can catch up once on wake. Better models can strengthen the method; the job and its boundaries stay visible.
A working life crosses tools, models and years. Give its knowledge a durable home: owned files, visible methods and reversible upkeep. Make the document a place to return to, with the freedom to carry it elsewhere.
Your expertise lives in the questions you ask and the distinctions you notice. Let a useful method preserve some of that care. Richer domain methods and documents that cooperate are a future chapter: more ways for hard-won knowledge to travel.
The argument, sources and routine belong together.
Make the distinctions visible. Improve the method as you learn.
Adapt a method or an included example to your own files.
Keep the folder when you need its working context. Richer domain libraries and cooperating documents are future directions.
05 / The invitation
The first result is a demonstration. The second cycle is the test. Did the document stay useful? Was the change easy to understand? Could you recover? Did the routine ask for less of your attention?
Download Flow. Give one page a purpose, a bounded routine and a way to review the result. Try the next cycle. Make room for the project, question or possibility you keep postponing.
The first result shows what is possible.
Check the source, understand the change, make your decision.
The routine ran without manufacturing another thing to read.
A missing source leaves an explanation, not an invented answer.
The measure is your next return: useful work, clear changes, a way back and less attention demanded.
A PRACTICE WORTH RETURNING TO
Flow updates, Living Documents methods, and the AI Native Newsletter.