Gelişimİş

Async Writing That Replaces Meetings: The 5 Document Formats High-Output Teams Use

A person writing in a notebook at a sunny desk by a window, laptop and tea beside them

TL;DR. The organizations that rely most on async work use a few documents with fixed sections, a named decider and a fixed comment window. Python proposals need an Abstract of about 200 words and a Rejected Ideas section. Rust’s final comment period “lasts ten calendar days, so that it is open for at least 5 business days.” An IETF Last Call runs at least two weeks. Amazon replaced slides with “narratively structured six-page memos.” The speed case is more modest than often claimed: adults read English non-fiction at about 238 words per minute (Brysbaert, 2019) and audiobooks are spoken at 140-180, so reading is only about 1.3 to 1.7 times faster (CEOtudent calculation). The bigger gain is that many people can read one document at different times, and it leaves a record. This guide gives five copy-ready templates (decision doc, narrative pre-read, pitch, weekly update memo, incident review), a format selection table and a checklist. The CEO move: put the decision owner and the response deadline at the top of every document, in writing. The student move: treat each document as practice in thinking clearly, and reread it cold before you send it.

This is the how-to layer under the async-first work architecture. If you still need to show why meetings are worth cutting, start with the true cost of meetings. To decide whether a piece of work needs people in a room at all, use the sync-or-agent decision protocol.

Why does writing replace meetings, and where is the evidence thin?

Marc Brysbaert’s 2019 meta-analysis in the Journal of Memory and Language pooled 190 studies with 18,573 participants. It estimates “the average silent reading rate for adults in English is 238 words per minute (wpm) for non-fiction and 260 wpm for fiction.” Oral reading averages 183 wpm. For silent reading of English non-fiction “most adults fall in the range of 175 to 300 wpm.” The paper also says reading rates “are in line with maximum listening speed,” and notes that audiobooks are spoken at 140-180 wpm. So silent reading at 238 wpm is about 1.32 times as fast as speech at 180 wpm and 1.70 times as fast as speech at 140 wpm (CEOtudent calculation): a real gain, not a tenfold one.

His figures are also “for people who keep on task the whole time,” and rates in less constrained situations “are likely to be lower because of mind wandering and task interruptions.” The reading times below are therefore best-case figures.

The stronger reasons to write are structural:

  • Decoupling. A document is read when the reader has attention to spare, not when the calendar allows. The Rust process makes this explicit: its comment window is set to cover at least five business days so that “all stakeholders have a chance to lodge any final objections before a decision is reached.”
  • One writer, many readers. A meeting costs every attendee the same minutes. A document costs the writer hours once, and each reader minutes.
  • A record. Python’s PEP 1 notes that because PEPs are kept in a versioned repository, “their revision history is the historical record of the feature proposal.”
  • Thinking. Writing exposes gaps in an argument before readers find them; the evidence is in writing as thinking.

There is one serious counterweight. Yang and colleagues studied the emails, calendars, chats and calls of 61,182 US Microsoft employees in the first six months of 2020. They found that firm-wide remote work made collaboration networks “more static and siloed,” with “a decrease in synchronous communication and an increase in asynchronous communication” (Nature Human Behaviour, 2022; abstract only). Documents travel across teams only if someone sends them there, so the templates include a distribution line.

Which organizations already run on documents?

Each system below is described in its own primary text. The Amazon, Basecamp and Google texts are company-published; the Python, Rust and IETF texts are open governance documents.

System (publisher) Document Required parts, as published Timing rule, as published Who decides
Python (PEP 1) Python Enhancement Proposal Preamble, Abstract (~200 words), Motivation, Specification, Rationale, Rejected Ideas, Open Issues, among others Title limited to 44 characters Steering Council or a PEP-Delegate
Rust (RFC process README) Request for Comments Summary, Motivation, Guide-level and Reference-level explanation, Drawbacks, Rationale and alternatives, Prior art, Unresolved questions, Future possibilities Final comment period of ten calendar days, at least 5 business days Sub-team; all members sign off before the final comment period
IETF (RFC 2026, RFC 7322) Internet-Draft and RFC Abstract that is “complete in itself” Draft public for not less than two weeks; Last-Call no shorter than two weeks, or four weeks if not started by a working group IESG
Amazon (2017 shareholder letter, SEC exhibit) Six-page narrative memo “Narratively structured”; no slides Read silently at the start of the meeting; a great memo “probably should take a week or more” to write Not specified in the letter
Basecamp (Shape Up, chapter 6) Pitch Problem, Appetite, Solution, Rabbit holes, No-gos Appetite stated as six weeks, or two weeks for a small batch The people who schedule projects
Google SRE (SRE book, chapter 15 and appendix D) Postmortem Impact, actions taken, root causes, follow-up actions; the example adds trigger, detection, lessons learned and timeline Written after defined triggers; senior engineers review the draft Review sessions finalize it; action items have named owners

Sources: PEP 1; Rust RFCs README and 0000-template.md; RFC 2026 sections 6.1.1 and 6.1.2; RFC 7322 section 4.3; Amazon.com 2017 letter to shareholders, Exhibit 99.1 to Form 8-K filed 18 April 2018; Shape Up chapter 6; Site Reliability Engineering, chapter 15 and appendix D. Figures as printed.

Three patterns repeat: fixed sections, so readers know where to look; a named decider, so comments feed a decision; and a timing rule, so silence after the deadline means something. GitLab’s handbook adds a fourth pattern for meetings that do still happen: agendas shared “72 hours (minimum 24 hours)” in advance, and default meeting lengths of 25 or 50 minutes.

How long does each format take to read?

The table applies Brysbaert’s rates to the upper word budget of each format. The budgets come from the CEOtudent framework below. The 25-minute slot is one of GitLab’s “speedy meetings” lengths.

Format Upper word budget Minutes to read at 238 wpm Minutes to read at 175 wpm Minutes to hear at 180 wpm Minutes to hear at 140 wpm Reading time as share of a 25-minute meeting
Weekly update memo 300 1.3 1.7 1.7 2.1 5%
Pitch 1,000 4.2 5.7 5.6 7.1 17%
Decision doc 1,200 5.0 6.9 6.7 8.6 20%
Incident review 1,500 6.3 8.6 8.3 10.7 25%
Narrative pre-read 3,000 12.6 17.1 16.7 21.4 50%

CEOtudent analysis of Brysbaert (2019) reading and speech rates and GitLab’s 25-minute meeting default. Word budgets are a CEOtudent editorial framework, not measured data. Real reading is slower when readers are interrupted.

Two things follow. A reader at 175 wpm takes about as long as a listener at 180 wpm, so for slower readers the case for documents rests on decoupling and the record, not speed. And budgets matter: a 3,000-word memo fills half of a 25-minute meeting, which is why Amazon builds reading time into the meeting itself.

Which format replaces which meeting?

Format Replaces this meeting Word budget Who reads Decision owner Response deadline
Decision doc Decision meeting, approval meeting 600-1,200 People affected by the decision, plus anyone with a veto One named person 5 business days for reversible decisions; 10 for hard-to-reverse ones
Narrative pre-read Slide presentation, quarterly or project review 1,500-3,000 Meeting attendees only Meeting owner Read before the meeting or in its first minutes; questions in the margin
Pitch Kickoff, brainstorm, scoping meeting 500-1,000 Whoever allocates time and budget The person who funds the bet Next planning cycle
Weekly update memo Status meeting, recurring check-in 150-300 The team and its direct stakeholders None; it informs, and links to decisions No reply needed; named asks by the date given
Incident review Blame meeting, ad hoc post-incident call 800-1,500 The team, then a wider engineering or operations audience Reviewers who close it out; one owner per action item Draft within 5 business days; comments closed before the review session

CEOtudent editorial framework. The comment windows are adapted from the Rust final comment period (at least 5 business days) and the IETF Last-Call minimums (two and four weeks). Budgets and deadlines are editorial choices for teams to adjust, not research findings.

If a meeting fits none of these rows, it is probably one of the meetings that should stay. The section on limits below lists them.

Format 1: How do you write a decision doc?

A decision doc replaces the meeting where people gather to agree. It borrows from three sources. PEP 1 requires that an author document “dissenting opinions” and record rejected ideas so people avoid “bringing up the same rejected idea again.” RFC 7282 says “lack of disagreement is more important than agreement,” and that rough consensus means issues are “addressed, but not necessarily accommodated.” Amazon’s 2016 letter says most decisions “should probably be made with somewhere around 70% of the information you wish you had,” and separates reversible “two-way doors” from decisions that are hard to undo.

DECISION DOC: [the decision in one line, 12 words or fewer]
Decision owner: [one name]          Status: [Open for comment | Decided | Withdrawn]
Comment deadline: [date, time, time zone]     Decision date: [date]
Reversibility: [Two-way door: easy to undo | One-way door: hard to undo]
Distribution: [teams and people who must see this, including other teams]

1. The ask (3 sentences maximum)
   We will decide [X] by [date]. The owner recommends [option].
   No objection by the deadline counts as agreement.
2. Context: [what changed, with numbers and a link to the data]
3. Options considered
   A. [option]: benefit [ ], cost [ ], main risk [ ]
   B. [option]: benefit [ ], cost [ ], main risk [ ]
   C. Do nothing: benefit [ ], cost [ ], main risk [ ]
4. Recommendation and why it beats the other options: [ ]
5. What we do not know yet (roughly what share of the information we would like): [ ]
6. Rejected ideas: [idea], rejected because [reason]
7. Dissent log: [name]: [objection] | [how it was addressed, or why it was not accommodated]
8. Decision record (owner fills in on the decision date):
   Decided: [option]   By: [owner]   Date: [ ]   Review on: [date]

A reader who stops after section 1 should still know what is being decided, by whom and by when. For decisions worth learning from, copy section 8 into a personal decision journal.

Format 2: How do you write a narrative pre-read?

The pre-read does not remove the meeting. It removes the presentation. Amazon’s 2017 letter describes it: “We don’t do PowerPoint (or any other slide-oriented) presentations at Amazon. Instead, we write narratively structured six-page memos.” The same letter says the quality of memos “varies widely.” It also says great ones are “written and re-written, shared with colleagues who are asked to improve the work, set aside for a couple of days, and then edited again with a fresh mind.”

PRE-READ: [topic]
Author: [ ]   Meeting: [date, time]   Reading time: about [words / 238] minutes
This meeting must answer:
  1. [question]
  2. [question]

1. Summary (150 words, complete in itself): [the whole argument in miniature]
2. Where we are: [facts, numbers and trend, with sources]
3. What we learned: [three findings, each with its evidence]
4. What we propose: [plan, owners, dates]
5. What could go wrong: [risks, and what would make us stop]
6. What we need from each reader: [name]: [specific question]
Appendix (optional reading): [tables, raw data, background]

The summary rule comes from the RFC style guide, which says an Abstract “should be complete in itself” because it often appears on its own. Readers who skim should be able to stop after section 1. Section 5 can be filled using the pre-mortem protocol rather than rebuilt here.

Format 3: How do you write a pitch?

A pitch replaces the open-ended kickoff or brainstorm. Basecamp’s Shape Up lists “five ingredients that we always want to include in a pitch”: Problem, Appetite, Solution, Rabbit holes and No-gos. Appetite is the part most proposals leave out. Shape Up describes it as “how much time we want to spend and how that constrains the solution,” and says that “stating the appetite in the pitch prevents unproductive conversations.”

PITCH: [project name]
Funding requested from: [person or group]   Decision by: [date]
Problem: [one specific story of the current way failing, for a specific user]
Appetite: [the most time we will spend, e.g. 2 weeks or 6 weeks; the scope
          shrinks to fit the time, the time does not grow to fit the scope]
Solution: [the core elements, as a short list or a rough sketch]
Rabbit holes: [details that could eat the time, and how we avoid each one]
No-gos: [what is deliberately out of scope]
What happens if we do not fund this: [ ]

The appetite line changes the question from “is this a good idea?” to “is it worth this much time?”

Format 4: How do you write an async status update?

This is the format most teams mean when they ask how to write an async update. It replaces the recurring status meeting. Shape Up gives the main design principle: managers dislike asking for status, and teams hide uncertainty because “nobody wants to raise their hand to management and say” they do not know how to solve a problem. Its answer is to report “what’s unknown and what’s solved” rather than counting tasks, using the uphill and downhill phases of work.

[Team or project] update, week of [date]
Status: [On track | At risk | Off track]   Author: [ ]   Next update: [date]

Bottom line (2 sentences): [the one thing a reader must know this week]

Shipped since last update (outcomes, not activity):
  - [what changed for a user or a number]
Next, until [date]:
  - [ ]
Uphill (still working out the approach):
  - [item]: the open question is [ ]
Downhill (approach known, executing):
  - [item]: expected done [date]
Blocked, or needed from a specific person:
  - [name]: [specific ask] by [date]
Decisions taken this week: [decision doc titles]
Metric: [name]: [value] vs target [value], trend [up | flat | down]

Keep it within 150 to 300 words and publish it on the same day each week in the same channel. The uphill line plays the role Shape Up gives a dot on its hill chart that does not move: “effectively a raised hand.” An item uphill for two updates in a row is the trigger for a short meeting.

Format 5: How do you write an incident review?

The SRE book defines a postmortem as “a written record of an incident, its impact, the actions taken to mitigate or resolve it, the root cause(s), and the follow-up actions to prevent the incident from recurring.” It must be blameless: it “must focus on identifying the contributing causes of the incident without indicting any individual or team for bad or inappropriate behavior.” The sections below follow the book’s example postmortem.

INCIDENT REVIEW: [short title] (#[id])
Date: [ ]   Authors: [ ]   Status: [Draft | In review | Complete]
Summary: [what happened, in 2 sentences]
Impact: [who was affected, how many, for how long]
Root causes: [contributing causes in systems and processes, not people]
Trigger: [what set it off]
Resolution: [what stopped it]
Detection: [how it was noticed, and how long that took]
Action items:
  - [action] | type: [mitigate | prevent | process] | owner: [name] | due: [date]
Lessons learned:
  What went well: [ ]
  What went wrong: [ ]
  Where we got lucky: [ ]
Timeline: [time, time zone]: [event]
Reviewers: [names]   Comments close: [date]   Shared with: [audience]

“Where we got lucky” is the section teams most often skip and the one that records near misses.

How do you write an async update people actually read?

Apply this checklist to all five formats before you send:

  1. The first two sentences carry the message. A reader who stops there knows the status, the ask and the deadline.
  2. The deadline has a date, a time and a time zone. “End of week” means different things in different places.
  3. One person is named as the decider or the owner. A document with no owner becomes a discussion thread.
  4. It stays inside its word budget. Anything over the budget moves to an appendix or a link.
  5. It reports outcomes, not activity. “Checkout errors fell” beats “worked on checkout.”
  6. Unknowns are separated from knowns. Uncertainty that is written down can be helped; hidden uncertainty piles up as risk.
  7. Every ask names a person. “Can someone look at this?” gets no answer; “[Name], can you approve by Thursday 15:00 UTC?” does.
  8. There is one source of truth. GitLab’s handbook warns that having “more than one agenda” is as bad as having none. Link to the canonical document instead of pasting copies.
  9. The template does not change. Readers learn where to look, so the same headings every week save them time.
  10. It has been reread cold. Amazon’s best memos are “set aside for a couple of days.” For a weekly update, an hour away from it is enough.

To check that the system works, count recurring meetings cancelled because a document did their job. GitLab’s handbook goes further, saying “roughly 1% of meetings should end within the first minute” when the outcome or preparation is missing.

How do the CEO and the student each use these documents?

The CEO move is ownership in writing. Every document above has an owner line and a deadline line at the top. Writing “Decision owner: [name], decided by [date]” settles what many meetings exist to settle, and makes silence meaningful. The Rust process does not require “consensus amongst all participants,” which it calls “usually impossible,” but it does require that the argument be “clearly articulated” before the decision window opens. Amazon’s 2016 letter adds “disagree and commit,” a way to move forward without unanimous agreement.

The student move is practice. Each document is a short, repeatable exercise in clear thinking: state the problem, weigh the options, name what is unknown. Bezos frames memo quality as a matter of scope rather than talent. Writers “mistakenly believe a high-standards, six-page memo can be written in one or two days or even a few hours, when really it might take a week or more.” Reread your own documents after three months; the rejected-ideas and dissent sections show where judgment can improve.

What this does not mean

Writing does not abolish meetings, and the systems above do not claim it does:

  • Postmortems still end in a meeting. The SRE book encourages “regular review sessions for postmortems” to “close out any ongoing discussions and comments.” Under its rule “no postmortem left unreviewed,” the document comes first and the review session second.
  • Amazon’s memo is read inside the meeting, and Rust still schedules calls. The memo replaces slides, not discussion. The Rust README notes that the sub-team “may schedule meetings with the author and/or relevant stakeholders,” with a summary posted back to the thread.
  • Some work needs real-time contact. Conflict, bad news about a person’s job, building trust in a new team and fast-moving incidents in progress need voice and faces. A recorded and summarized meeting is not a substitute either; see what AI meeting notes changed.
  • Async can isolate teams. The Microsoft study found networks became “more static and siloed” as communication moved to async channels. Distribution lines and cross-team readers are part of the design, not extras.
  • Documents cost writing time. A 1,200-word decision doc that takes 5 minutes to read can take hours to write. For a two-person question, a short call may be better.

Frequently asked questions

How long should an async status update be?
Within 150 to 300 words in this framework; 300 words take about 1.3 minutes at 238 words per minute. Move detail to links.

Is reading really faster than a meeting?
Somewhat: about 1.3 to 1.7 times as fast as audiobook speech, and close to even for slower readers. The main gains are timing, reach and the record.

What is the difference between a decision doc and a pitch?
A pitch asks for time and money to explore a solution within a fixed appetite. A decision doc asks people to commit to one option from several, by a deadline, with a named owner.

How long should a comment window be?
This framework uses 5 business days for reversible decisions and 10 for hard-to-reverse ones. That is adapted from Rust’s ten-calendar-day final comment period and the IETF’s two-week Last-Call minimum. The exact length matters less than writing it down at the top.

Does no response mean agreement?
Only if the document says so before the deadline. RFC 7282 argues that “lack of disagreement is more important than agreement,” but that only works when everyone knows the rule. Put it in the ask section so nobody is surprised.

Should AI write these documents?
AI can draft the context and timeline sections. The ask, the recommendation, the rejected ideas and the unknowns are where the author’s own thinking is the point of the document. Draft those yourself first.

Sources

  1. Brysbaert, M. (2019). How many words do we read per minute? A review and meta-analysis of reading rate. Journal of Memory and Language, 109, 104047. (author manuscript version)
  2. Amazon.com, Inc. (2018). 2017 Letter to Shareholders. Exhibit 99.1 to Form 8-K filed 18 April 2018, SEC EDGAR. (company-published)
  3. Amazon.com, Inc. (2017). 2016 Letter to Shareholders. Exhibit 99.1 to Form 8-K filed 12 April 2017, SEC EDGAR. (company-published)
  4. Warsaw, B., Hylton, J., Goodger, D. and Coghlan, A. PEP 1: PEP Purpose and Guidelines (created 2000, Active). Python Software Foundation, peps.python.org.
  5. The Rust Project. Rust RFCs README (RFC process) and 0000-template.md, rust-lang/rfcs repository.
  6. Bradner, S. (1996). RFC 2026: The Internet Standards Process, Revision 3. IETF, BCP 9.
  7. Flanagan, H. and Ginoza, S. (2014). RFC 7322: RFC Style Guide. RFC Editor.
  8. Resnick, P. (2014). RFC 7282: On Consensus and Humming in the IETF. IETF.
  9. Singer, R. Shape Up: Stop Running in Circles and Ship Work that Matters, chapter 6 (Write the Pitch) and chapter 13 (Show Progress). 37signals, free online edition. (company-published book)
  10. Lunney, J. and Lueder, S. Postmortem Culture: Learning from Failure (chapter 15) and Example Postmortem (appendix D). In Site Reliability Engineering: How Google Runs Production Systems. Google and O’Reilly Media, 2017 online edition (CC BY-NC-ND 4.0).
  11. GitLab Inc. GitLab Handbook: All-Remote Meetings; Communication (asynchronous communication section). (company-published)
  12. Yang, L., Holtz, D., Jaffe, S., Suri, S. and others (2022). The effects of remote work on collaboration among information workers. Nature Human Behaviour, 6, 43-54. (abstract only; the journal page lists an author correction published 5 October 2021)

Note on method: the reading-time table is a CEOtudent analysis of Brysbaert (2019) reading and speech rates applied to the CEOtudent word budgets and GitLab’s 25-minute meeting length. The format selection table, the five templates, the word budgets, the comment windows and the checklist are a CEOtudent editorial framework adapted from the primary documents listed; they are not research findings. The table of document systems reports requirements as printed in each source. CEOtudent ran no survey. Deliberately not used: the claim that people read several times faster than they listen and the 300 wpm benchmark (Brysbaert’s review argues both overstate reading speed), and widely repeated survey percentages on unproductive meetings, whose primary texts could not be opened in full.


This content was compiled with the support of AI following in-depth research, then written and prepared for publication by the CEOtudent editorial team.

This post is also available in: Türkçe Français Español Deutsch

Benzer içerikler