Synthetic Interview Transcripts

[00:00]

Interviewer: So, tell me about how you write on your computer day to day.

Sarah: Oh god, Gmail. Like, embarrassingly so. I would say 60-70% of my day is in Gmail, then Slack, then Salesforce. And then probably Notion for internal stuff. But really it's email. My team jokes that I'm part email.

Interviewer: That's a lot of email. What kind of emails are you writing mostly?

Sarah: Everything. Follow-ups with prospects, internal updates, deal reviews... I probably write 15-20 versions of "thanks for your time, here's what we discussed" per week. And every time I'm like, didn't I just write this? But it's never exactly the same, so I can't just copy-paste.

[00:02:15]

Interviewer: What frustrates you about typing the same things over and over?

Sarah: The time, obviously. But also the mental drain? Like, I know what I want to say, and getting it out is just... friction. It's not creative work, it's not strategic work, it's just — production. And I'm doing it at 10pm because I didn't get to it during the day.

[00:04:30]

Interviewer: Have you tried any AI writing tools or autocomplete?

Sarah: I used Grammarly for like a week. It was fine for typos, I guess, but the tone suggestions were weird. It kept wanting to make me sound more formal. And I'm in sales, I need to sound human.

[00:06:45]

Interviewer: How do you feel about AI tools that send your writing to the cloud?

Sarah: For my personal email, I don't care that much. But for work? A lot of what I write involves deal terms, pricing discussions, competitive intelligence. If that's going to some server somewhere... I'd need to really trust them. And I don't trust startups with my deal data.

Interviewer: What if it ran entirely on your machine?

Sarah: Wait, that's possible? Like, the AI is just on my laptop? I didn't know that was a thing. Yeah, that changes the calculus completely. If nothing leaves my machine, I'd at least try it.

[00:09:00]

Interviewer: What would matter most to you?

Sarah: Accuracy, hands down. If it's suggesting wrong words in a client email, that's worse than no suggestion. It has to know my voice. Speed matters too — if there's a delay, I'll just keep typing. And it has to work in Gmail. In the actual compose window. Not some weird sidebar.

[00:12:00]

Interviewer: Would you pay for this?

Sarah: If it saves me even 30 minutes a day, that's worth... a lot. I'd pay $20/month easily. Actually, I'd prefer a one-time purchase. I hate subscriptions. If you told me $150 once and it's mine? I'd probably buy it today.

[00:14:30]

Interviewer: What would make you stop using it?

Sarah: Being wrong in front of a client. If it suggests something and I don't catch it and I send it... that's my reputation. Also, if it slows me down. And if it doesn't work in Gmail — I said this already but I really mean it — if I have to switch to some other app to use it, it's dead to me.

[00:00]

Interviewer: Tell me about how you write on your computer day to day.

Marcus: VS Code, obviously. But I'm also in Slack constantly — like, embarrassingly constantly. And then GitHub for PRs and reviews, Linear for project management... I write a lot more English than people probably assume. Like, I probably spend two hours a day just writing in Slack.

[00:02:00]

Interviewer: What frustrates you about typing the same things?

Marcus: Slack, man. Every morning I write like "hey, blocked on X, need Y from Z" and then three variations of that throughout the day. It's not a huge deal, but it's... it's low-grade annoying. Like a squeaky door.

[00:04:00]

Interviewer: How has Copilot been for code?

Marcus: Oh, Copilot's great for code. Like, genuinely great. But here's the thing — it only works in my editor. The second I'm in Slack or a browser, it's gone. And I write a LOT of non-code.

[00:06:30]

Interviewer: Cloud vs local?

Marcus: For code, I've accepted it. But for Slack? For PR reviews where I'm talking about internal stuff? I'd prefer it was local.

Interviewer: Do you think a local model could be good enough?

Marcus: That's the question, isn't it? For basic sentence completion? Probably. But for context — like, understanding a whole Slack thread — I don't know. Apple Silicon is pretty beefy now though.

[00:09:00]

Interviewer: What would matter most?

Marcus: Latency. If there's any perceptible delay, I won't use it. With Copilot, if the suggestion takes more than like 200ms, I've already typed the next word. And it needs to work in Slack specifically. Inline, where I'm typing.

[00:12:00]

Interviewer: What would stop you?

Marcus: If it's wrong too often. If it's right 70% of the time, I'll probably turn it off because the 30% wrong is more annoying than typing everything myself. Also, if it breaks Alfred snippets or Karabiner mappings — that's a dealbreaker. My keyboard setup is sacred. And if it doesn't understand developer context — if it suggests "that's a great point!" instead of something technical, I'm uninstalling.

Interviewer: Anything else?

Marcus: I'd want it to be keyboard-dismissable. Like, one key to dismiss, no mouse. That's just table stakes for a developer tool.

[00:00]

Anika: Google Docs, mostly. I'm writing papers all the time. Then email (Gmail), Slack for the lab group, and Overleaf for LaTeX papers.

[00:02:30]

Interviewer: What frustrates you about writing in English?

Anika: There's a difference between being fluent and being native. Sometimes I'll write something and it's grammatically correct but it doesn't sound quite right. And this matters in my field! Reviewers will sometimes comment on the language. That's humiliating.

[00:05:30]

Interviewer: What tools do you use?

Anika: Grammarly. I have a love-hate relationship with Grammarly. It's always telling me to be more concise. But in scientific writing, sometimes a long sentence is necessary! The tone suggestions are so condescending. "This sounds aggressive." No, it doesn't, it sounds precise!

[00:08:00]

Interviewer: What would matter most?

Anika: That it works in Google Docs and Overleaf. Also, I don't want it to be too eager. It should know when to help and when to stay out of the way.

[00:10:00]

Interviewer: Would you pay?

Anika: A one-time purchase of $50-80? That I'd do. But it has to learn my technical vocabulary. If it's just a generic English improver, Grammarly already does that for free.

[00:12:30]

Interviewer: What would stop you?

Anika: If it made my writing sound like everyone else's. I've worked hard to develop my academic voice. I want assistance, not erasure. Also, if it's constantly suggesting changes, it would make me anxious. I already deal with imposter syndrome about my English.

[00:00]

Priya: Google Docs for scripts — 200+ docs in my "Scripts 2026" folder. Notion for planning. Gmail for sponsors. On a writing day, 8,000 to 12,000 words.

[00:02:30]

Interviewer: What frustrates you?

Priya: The transitions. The intros. The sign-offs. The "hey everyone, welcome back" that I write 150 times a year. And the sponsor reads! I have to write a fresh ad read for every video. That takes 45 minutes each.

[00:05:30]

Interviewer: You've tried AI tools?

Priya: I've tried Jasper, Copy.ai, ChatGPT. They're great for first draft and then terrible for everything after. My audience can tell. I tried an AI-written sponsor read once and someone commented "did an AI write this?" I never did it again.

[00:08:00]

Interviewer: Would you pay?

Priya: I pay for Grammarly Premium, Jasper, eight subscriptions. For a tool that saves me 30 min/day on scripts, $15/month easy. One-time? Even better. $200 once? I'd buy it right now.

[00:10:30]

Interviewer: What would stop you?

Priya: If my audience could tell I was using it. That's number one. If it slows down Google Docs. If it requires a lot of setup. And I'd want to turn it off for personal stories — some parts should just be me.

[00:00]

Rachel: Figma, Slack, Linear, Notion, Gmail, iMessage, Twitter. I'm a chronic over-communicator. My team makes fun of me for sending three-paragraph Slack messages when "ok" would suffice.

[00:02:30]

Rachel: I'm on Espanso, 150 snippets. It's genuinely life-changing. But it's static. If I'm writing to a client vs my team, the same snippet doesn't work. I end up with 30 variations and can't remember which one to use.

[00:05:30]

Rachel: Apple Intelligence writing tools were absolutely terrible. The suggestions appeared in a floating panel in the wrong place. I also tried Compose AI — fine in Gmail but nowhere else. I want something at the OS level.

[00:08:30]

Interviewer: Would you pay?

Rachel: One-time purchase, absolutely. $100-150, easy. Monthly? Maybe $5/month. I'm at my limit for subscriptions.

Interviewer: What would stop you?

Rachel: Bad UX. Slow performance. Any visual jank — flickering, wrong positioning. I'm a UX designer, I will notice. And I will judge. Also: if it doesn't work in Electron apps. I live in Slack, Figma, Linear — all Electron.

[00:00]

Daniel: Google Docs and Confluence mostly. API references, how-to guides, internal wikis. Then Slack and GitHub for PR descriptions.

[00:02:30]

Daniel: Text expansion keeps me sane. ":apiref" gives me a template. ":prdesc" for PR descriptions. I've been building this library for five years. But release notes — same structure, different content every time. If AI could look at an engineer's Slack message and draft a release note in our style guide's voice? Incredible.

[00:05:30]

Daniel: Local AI is really appealing. I write about internal systems, APIs, roadmaps. That stuff is confidential. My company has a policy against ChatGPT now.

[00:08:30]

Daniel: I don't want it to be creative. I want it to be helpful. Boring is good in documentation. Better to suggest nothing than to suggest something wrong — accuracy is non-negotiable. Also, don't mess with my Espanso snippets.

[00:00]

Emily: Google Docs, for sure. Proposals and reports. Gmail for client communication. I write probably two or three proposals a week, each takes three to four hours. The actual writing part is an hour and a half, and a lot of it is stuff I've written before.

[00:02:30]

Emily: I'm a consultant, right? My time is literally what I sell. If I could cut proposal writing time in half, that changes my business. I have 80 past proposals and I search through them manually.

[00:05:00]

Interviewer: Privacy concerns?

Emily: My proposals contain client strategies, financial projections, competitive analysis. I sign NDAs. If it's on my computer and nothing leaves, that's better. But honestly, I don't fully understand how any of this works.

[00:07:30]

Interviewer: What would stop you?

Emily: If it was complicated to set up. If it didn't work in Google Docs. If it made mistakes I had to catch. And if it changed my voice — my proposals win contracts because they sound like me.

[00:00]

Alex: iA Writer for first drafts — the only writing app that gets out of your way. Then Google Docs for client collaboration. I spend 45 minutes writing an email to a client saying "here's the first draft, let me know your thoughts." That email does not need my full creative power.

[00:02:00]

Alex: I have OPINIONS about AI writing tools. GrammarlyGO was competent but lifeless. Jasper writes like a content marketer from 2019 — "In today's fast-paced world." Apple Intelligence was patronizing. They all try to write FOR me instead of WITH me.

[00:05:30]

Alex: The ideal tool would suggest the word I'm about to type. Not rewrite my sentence. Not generate a paragraph. Just the next word. Small suggestions that keep my flow going.

[00:08:00]

Interviewer: Would you pay?

Alex: One-time, yes. $150, maybe $200. Subscription, probably not. I'd be skeptical of a subscription for a local tool — what am I paying for monthly if the model runs on my machine?

Interviewer: What would stop you?

Alex: If the suggestions were visible to the client during screen share — that's embarrassing. Also, if it doesn't work in iA Writer. That's my primary tool.

Interviewer: Do you think AI autocomplete threatens professional writers?

Alex: No. Autocomplete is not writing. It's typing assistance. The value of a professional writer isn't in producing words — it's in knowing which words, in what order, to achieve a specific effect. But an AI that helps produce those words faster? That's just a better keyboard.