Eight things to actually do with it on Tuesday

1. 1

   Load the dictionary with the course, once per term

   The lecturer's name, the module code, the eight terms that will be on every slide. This is the highest-leverage two minutes in the whole workflow, because the dictionary is sent with the audio — the recogniser is already leaning toward those words while it decides what it heard, instead of you find-and-replacing a mangled term out of forty pages of notes in April.

2. 2

   Record the hour and stop thinking about it

   In the room, the mic is enough. If the lecture is remote, the meeting recorder takes mic and system audio together, and nothing joins the call to announce that you are recording it.

3. 3

   Let the labels land

   Speaker separation splits the lecturer from the questions at the end — which is usually the ten minutes where what is actually examinable gets said out loud.

4. 4

   Make the flashcards the same evening

   Turn the note into a spaced-repetition set while you still remember which bits were confusing. Reviewing on a schedule is the part that converts an hour of attendance into something you can still produce under exam conditions.

5. 5

   Revise across the whole term, not note by note

   In the week before the exam, ask AI chat a question of everything you have recorded at once. "Where did she contradict the textbook" beats opening eleven recordings and scrubbing each one.

The lecture stops being an hour of audio you will never replay, and becomes the only revision material you did not have to write.

1. 1

   Walk and talk about the one thing you have an opinion on this week

   Do not structure it. Contradict yourself, backtrack, start the sentence again. The whole reason this beats sitting at a document is that a walk removes the pressure to be publishable in the first draft — and the cleanup exists precisely so that the mess costs you nothing.

2. 2

   Set Auto Cleanup to Medium or High before you look at it

   Twenty minutes of thinking out loud arrives as something with paragraphs. If the rewrite flattened a line you liked, "what I actually said" gives you the raw version back and you take that sentence out of it by hand.

3. 3

   Pull the short post out with Make it shorter

   Run it on the section that carries the argument, not on the whole recording. Twenty minutes compressed uniformly turns into an outline; one paragraph compressed hard turns into a post.

4. 4

   Run your own transform for the rest of the week

   Write a custom transform once — "split this into five posts, each with its own opening line, no hashtags" — and it is available on every recording after it. This is the step that makes the walk worth doing weekly rather than once.

5. 5

   Make the visual in the same app

   Image studio for the thing that goes next to the words, so the post is not sitting in drafts for three days waiting on a picture.

One walk, a week of drafts, and not a single blank page in the sequence.

1. 1

   Turn your repeated paragraphs into Snippets

   The refund policy wording, the how-to-export walkthrough, the escalation caveat, the sign-off. Each gets a spoken cue. You are not saving keystrokes here so much as saving the version drift that happens when eleven people paraphrase the same policy from memory.

2. 2

   Pin the product nouns the recogniser keeps mangling

   Plan names, feature names, the integration whose spelling nobody agrees on. Pinning means they survive; the dictionary also learns from the corrections you make, so the list gets better without you maintaining it.

3. 3

   Answer the ticket out loud, as if the customer were across the desk

   Hold the Flow Bar and explain it. Speak the cue where the boilerplate goes and the Snippet expands mid-sentence. Talking produces a warmer answer than typing does, which is the part nobody expects — the written version of you is more clipped than the spoken one.

4. 4

   Turn into a reply, then Make it shorter

   The first transform gives it the shape of a reply rather than a monologue. The second one is for when you have over-explained, which after four hours in a queue you reliably have.

5. 5

   Land it in the box

   On the desktop build the Flow Bar is system-wide, so the text arrives in the reply field you already have open. On the web or a phone you copy it out of Waver. Either way there is no integration to configure and nothing that needs access to your helpdesk.

The queue moves at talking speed, and the fortieth answer of the day is as careful as the first.

1. 1

   Ask your own notes what you agreed last time

   AI chat across everything you have recorded, rather than scrolling back through a month looking for the commitment you half remember making. Walking in with the previous conversation actually loaded is most of what people mean when they say a manager is good at 1:1s.

2. 2

   Record it, and say that you are

   The meeting recorder takes mic and system audio with no bot joining the call — which means nothing announces it on your behalf, so announce it yourself. A 1:1 is exactly the conversation where being recorded without being told poisons the well.

3. 3

   Do not take notes

   This is the actual change. Note-taking during a 1:1 splits your attention at precisely the moment someone is deciding whether to tell you the real thing. The recording is doing that job.

4. 4

   Write the follow-up off the labelled transcript

   Speaker labels arrive after the meeting, so it is unambiguous who committed to what. Turn into a reply gives you the message you owe them; send it while the conversation is still an hour old.

5. 5

   Decide what persists

   Privacy Mode and Cloud Sync are two independent toggles, and with both set there is zero retention. Some 1:1s should leave a written trail. Some very much should not, and that is a decision you get to make per conversation rather than per company.

The follow-up goes out in the only hour it counts, and you spent the meeting listening.

1. 1

   Pin your stack in the dictionary

   Repo names, internal service names, and the lowercase tools — kubectl, pnpm, ripgrep — that no amount of context would let a recogniser guess. These are the words that make dictated technical prose unusable when they come out phonetically, and they are a one-time fix.

2. 2

   Explain the change to an imaginary reviewer

   Hold the Flow Bar and say what broke, what you changed, and what you deliberately did not change. That third one is the part reviewers actually want and the part that never survives the trip to a text box, because typing it feels like admitting something.

3. 3

   Polish for the PR description

   You get the structured version — the one with the reasoning in it, because you said the reasoning out loud instead of compressing it into a title.

4. 4

   Make it shorter, on the same recording, for the channel

   The Slack update is the same explanation at a tenth of the length. Two audiences, one act of thinking, no second pass at the blank box.

5. 5

   Prompt engineer when the next reader is a model

   If the next step is handing context to a coding agent rather than a human, that transform shapes the same two minutes into something a model can act on.

The writing around the code stops being the reason a finished branch sits unopened until Thursday.

1. 1

   Talk out the three things you need from the call

   On the way in, out loud. Set Auto Cleanup to Light so it stays recognisably your words — prep notes are for you, and over-polishing them removes the blunt phrasing that was the point.

2. 2

   Ask your notes what they said last time

   AI chat across previous recordings surfaces the thing they mentioned in passing eight weeks ago. Remembering it in the call is worth more than any slide you could have made instead.

3. 3

   Record the call itself

   Mic and system audio together, no bot in the participant list — which matters more with clients than internally, because an unexplained attendee named after a note-taking product is a conversation you then have to have.

4. 4

   Split what they committed to from what you did

   Speaker separation makes that unambiguous after the fact. Nearly every billing dispute is a disagreement about which side of that line a sentence fell on.

5. 5

   Two documents from one recording

   Turn into a reply for the recap email that goes to them. A custom transform you wrote once — "internal note: risks, scope creep, what they did not say" — for the version that does not.

The write-up happens while you still remember what they meant, not just what they said.

1. 1

   Capture it badly, in forty seconds

   Auto Cleanup off or on Light. You want the raw thought, including the bit where you are not sure yet — a tidied version of a half-formed idea reads more finished than it is, and you will trust it more than it deserves.

2. 2

   Add to it whenever it resurfaces

   On a walk, in a queue, between two meetings. Ideas do not arrive whole; they arrive four times in a fortnight, and the reason most of them die is that no two of those four occasions produced a written trace.

3. 3

   Pull the fragments together

   AI chat across your notes gathers the scattered takes into one place, and will tell you what is missing when you ask it what a reader would still not understand.

4. 4

   Run the spec transform

   A custom transform — "one page: problem, proposal, non-goals, open questions" — written once and reused forever. The structure is the part you never feel like doing at eleven at night, which is exactly when the idea is fresh.

5. 5

   Prompt engineer if the next reader is a model

   Same recording, different shape, for when the spec goes to a coding agent before it goes to a person.

Ideas that used to evaporate somewhere near the towel rack turn into the document you send on Monday morning.

1. 1

   Just talk, in whichever language the thought arrived in

   Language is detected automatically across 120+ of them and written in its own script. There is no settings trip to switch the input language and no second trip to switch it back — which is the friction that, in practice, makes people give up on dictation entirely when they work across two languages.

2. 2

   Keep both languages' names in one dictionary

   Colleagues, clients, places, product names. The dictionary rides your account rather than the device, so a term you add on the phone is there on the laptop and the tablet at the next recording.

3. 3

   Cross over when the other side has to read it

   Live translation is for the moment the output has to land in a language you do not draft in — a supplier, a client, a form. Your thinking does not have to happen in the language your job outputs.

4. 4

   Shape it for whoever is reading

   Transforms apply the same way regardless of which language you spoke — the reply that has to sound like a reply still sounds like one.

The tax you have been paying for working bilingually turns out to have been optional.