How to Keep the Decisions That Were Only Made Out Loud: A 3-Line Record for Engineers

design review notesarchitecture decision recordADR templateengineering meeting minutesspec decided verbally
How to Keep the Decisions That Were Only Made Out Loud: A 3-Line Record for Engineers

Near the end of a design review, fifteen minutes of back and forth ends in "let's go with that." All three people in the call are satisfied.

Two weeks later you sit down to implement it and stop. You cannot reconstruct why it was shaped that way. Scrolling back through Slack turns up a link to a doc and a message about the meeting time. The moment the decision was made exists nowhere. So the discussion happens again.
This is not an article about writing better minutes. It is about writing less and still not losing decisions, aimed at engineers who sit in design reviews and spec calls.

The short version

What disappears from a meeting is not the decision. It is the option you rejected and the reason you rejected it. Without that, nobody can explain the design six months later, and the same debate resurfaces.
Do thisWhy
Keep three fieldsDecision, assumptions, open. No transcripts
Record the meetingYou cannot type while presenting. Extraction can wait
Promote heavy decisions to an ADROne file per decision, with the reasoning attached

Full minutes fail because they are long. Three fields take five minutes after the call.

What is lost is the rejected option, not the decision

The decision survives. The reasoning does not.

"We went with token-based auth" is recoverable from the code. What vanishes is everything upstream of it:

  • Why sessions were ruled out
  • Which constraints forced the call (legacy systems, deadline, who operates it)
  • Who was uneasy about what
Six months on, a new teammate asks why the design looks like this and nobody can answer. Or someone proposes changing it and you cannot articulate why it is expensive to change. What has been lost is not a document. It is the ability to reproduce the judgment.

The cost is re-litigating the same call

When reasoning is not recorded, a rejected option comes back around. Nobody is being difficult; the constraint simply was not written down. So the team re-runs the analysis and lands in the same place.

Meeting records are the cheapest available defense against that kind of rework. Trim what you write and the cost approaches zero.

Three lines is enough

Decision, assumptions, open

After a meeting, these three fields carry the load:

## 2026-08-31 Payment retry design review

### Decision
- Exponential backoff, max 3 retries. Attempt 4+ goes to a manual queue

### Assumptions
- Payment API rate limit is 60/min (revisit if it changes)
- SRE handles retries from the manual queue by hand for now

### Open
- Where the manual queue UI lives (decide at the 9/7 review)
Assumptions matter more than they look. A decision lives exactly as long as its assumptions do. If the rate limit changes, this decision is back on the table. Writing them down lets whoever arrives later judge for themselves whether it still holds.

Do not skip the open items either. Things that were not decided disappear unless recorded. Naming the date they get decided keeps them from floating.

Decide what you are not writing

Minutes collapse under their own weight. You can leave all of this out:

  • Verbatim quotes (who phrased what rarely affects the decision)
  • The narrative of the discussion ("A proposed X, then B pointed out...")
  • Small talk and status updates (the tracker already has them)

There are moments when a verbatim record matters, and for those, keep the recording. You do not have to type it out.

Share it within 24 to 48 hours

Circulating notes within 24 to 48 hours of the meeting lets people correct misunderstandings while memory is fresh. Past that window, corrections stop arriving and whatever you wrote becomes the official version, mistakes included.

With three fields you can write it immediately after the call, which is also when it is easiest.

Promote hard-to-reverse decisions to an ADR

What an ADR is

An ADR (Architecture Decision Record) is a lightweight document holding one architectural decision and the reasoning behind it, one decision per file.

The minimal template, known as the Nygard format, has five parts:

FieldContent
TitleThe decision, stated briefly
StatusProposed / accepted / deprecated / superseded
ContextThe background and constraints that forced a decision
DecisionWhat was chosen
ConsequencesUpsides and downsides that follow

If you want the alternatives compared explicitly, the MADR 4.0 line of templates (published September 2024) structures the options side by side.

ADRs are not a personal habit. Microsoft's Azure Well-Architected Framework treats maintaining them as an architecture practice, which is useful ammunition when you propose the idea to a team.

Never edit one, supersede it

An ADR is not revised. When a decision is overturned, write a new ADR and mark the old one superseded.

That looks inconvenient, and it is deliberate. The path the reasoning took is itself information. Reading "we decided this, then changed it for this reason" is what helps the next person facing a similar call.

Not every meeting produces one

The test for whether something deserves an ADR is simple: is this decision hard to reverse later?
Kind of decisionWhere it goes
Auth scheme, datastore choice, API contract, async strategyADR
Library upgrades, naming conventions, retry count tuningThe decision field in your notes
Task assignments, review schedulingTracker and calendar

Almost everything from a weekly sync fits in three lines. One or two ADRs a month is already a lot. Without that filter, writing ADRs becomes theater.

Start from the fact that you cannot take notes during the meeting

Whoever is presenting has no hands free

There is a physical problem with everything above. In a design review, you are usually the one presenting.

You share your screen, walk through code, answer questions and sketch a diagram. Typing notes in that state is not possible. "Let's assign a notetaker" means, in a three-person review, removing one third of the discussion.

The predictable result: the meetings that produce the most decisions are the ones with the least record.

Record it, extract afterwards

Reverse the order. Stay in the discussion, and let the recording do the capturing. Afterwards, pull out only the three fields.

A recording keeps things an audio-only note would drop:

  • Diagrams drawn on a whiteboard or a canvas tool
  • Shared code and the diff on screen
  • Reproduction steps and the error output
  • What "this part, here" was actually pointing at

Design discussions rarely stand up as words alone. "Let's split this function" is unreadable later without the screen it referred to.

Doing this on a Mac

Qureco Screen Recorder, the macOS app we build, is aimed at exactly this pattern:
  • Screen and audio recording are free with no time limit, and no watermark
  • Your voice and the other side's audio are both captured with no virtual audio driver (no BlackHole setup)
  • On the Pro plan, AI generates minutes from the recording. Templates are configurable, so setting it to output "Decision / Assumptions / Open" cuts the extraction work further
  • Generated minutes can be sent straight to Notion

Pro is free for the first month, no credit card required.

One caveat: Qureco is macOS only, with no Windows or Linux version, and it does not show live captions during a call. It is built on the assumption that you use the recording after the meeting, not during it.

The reasoning behind capturing meetings without a bot joining is in AI meeting notes without inviting a bot, the transcription workflow in how to transcribe meetings automatically on a Mac, and turning decisions into tracked work in how to extract tasks from meeting minutes into Notion.

Say one sentence before you start recording. "I am recording this to write up the notes" is enough. Internal reviews almost never object, and stating it makes sharing easier afterwards.

Keep the pieces in different places

Three-field notes and ADRs work better stored separately.

WhatWhereWhy
ADRsIn the repo (docs/adr/0001-xxx.md)Reviewed and versioned alongside the code
Three-field notesNotion or your docs toolNon-engineers read them; you search them later
The recording itselfCloud, linked from the notesLarge files. Only some people need to open them
Then link them together: the ADR's context field points at the meeting notes, the notes point at the recording. The goal is being able to trace "why," not consolidating everything into one place.
Designing this for a whole team is covered in how to manage meeting minutes in Notion, and feeding the archive to an AI in can Notion MCP be used for meeting notes?.

Starting at tomorrow's review

  1. Say one sentence and start recording ("recording this for the notes")
  2. Stay in the discussion. Take no notes
  3. Afterwards, write the three fields (five minutes; open the recording only where needed)
  4. Share it the same day. Fix whatever people correct
  5. If a hard-to-reverse decision came up, write one ADR and link the meeting notes

For the first week, steps 3 and 4 are plenty. ADRs can wait until a decision actually calls for one.

FAQ

How do I tell people I am recording?

"I am recording this to write up the notes" covers it. With external attendees, ask once at the top of the call. The wider legal picture is in is it illegal to record a meeting?.

How is an ADR different from a design doc?

A design doc proposes and circulates the design of something you are about to build. An ADR records one decision and its reasoning. Work through a design doc, and when a significant decision falls out of it, split that into an ADR.

Who should write the notes?

With only three fields, you no longer need a designated notetaker. Accuracy is highest when whoever presented writes them afterwards. Given that a recording exists, "whoever noticed writes it" works better than a rotation.

Is an AI summary enough on its own?

AI is good at pulling out decisions, but assumptions and open items are what it tends to drop. Rather than filing the generated text as is, check that those two fields are populated. Specifying them in a template improves the odds.

How do I track the open items?

Write them in the notes, then create tickets for anything with a date. "We will decide next review" alone tends not to make it onto the next agenda.

Does this change for in-person meetings?

In a room there is no shared screen, so whiteboard photos have to go into the notes. Recording in-person discussions is covered in recording in-person meetings on a Mac.

What if I develop on Windows?

The three-field notes and the ADR practice are OS independent. Only the recording tool needs to be something that runs on Windows.

Glossary

TermMeaning
ADRArchitecture Decision Record. One architectural decision and its reasoning, one file each
Nygard formatThe minimal ADR template: title, status, context, decision, consequences
MADRAn ADR template that structures the compared options. The 4.0 line was published September 2024
SupersededStatus marking an old ADR replaced by a newer one rather than edited
Design docA document proposing the design of something to be built. Broader scope than an ADR
AssumptionA condition the decision depends on. If it changes, the decision is revisited
OpenSomething the meeting did not settle, recorded with the date it will be

Summary

  • What a meeting loses is not the decision but the rejected option and its reasoning, which is why the same debate returns
  • Keep three fields: decision, assumptions, open. Skip transcripts and narrative
  • Share within 24 to 48 hours so misunderstandings get corrected while memory is fresh
  • Promote only hard-to-reverse decisions to an ADR, and supersede rather than edit
  • In a design review you are presenting and have no hands free, so record it and extract the three fields afterwards

Minutes do not fail for lack of discipline. They fail because there is too much to write. Cut it to three fields, hand the rest to a recording, and the practice survives past tomorrow's review.

Qureco

Qureco Screen Recorder

Powerful screen recording app for Mac

Record meetings, let AI handle the notes, just read what arrives in Notion.Try all features free for the first month.

No Setup RequiredNo WatermarkAI Meeting NotesNotion Integration

About the Author

Shunsuke Inoue

Shunsuke Inoue

CEO, Qurio Inc.

Founder of Qurio, an AI consulting company. Majored in AI at Sophia University and founded the AI research circle "SOMA." As CEO of JPMT Inc., developed "MinPro" (1,300+ users) and business analysis SaaS "Optpath." Established Qurio Inc. in October 2025, focusing on AI and data development consulting. Speaker at the 30th Nikkei Forum "Future of Asia." Committed to promoting technological advancement and creating new value through AI.