In partnership with

The Rollout

Depending who you ask, somewhere between 12% and 44% of Australian businesses use AI.

Both numbers are real. The ABS says 12%. The National AI Centre says 44%. The gap is what counts as "using it."

Most of that 44% is someone with a chat tab open. The top two uses among SMEs are writing content and analysing data, 54% each. Useful. But it's a person driving a tool.

The 12% is closer to a system. Something that runs whether or not you're the one typing.

This week I built one of those on camera. Start to finish, about 20 minutes.

THE BUILD
Four parts. No code.

Here's the whole thing.

  • A folder of past proposals. That's the database. Ten in the demo. Three is enough to start.

  • A skill. A text file with the rules. What sections, what order, how we price, what we never quote. The stuff you'd tell a new sales rep on day one.

  • The discovery call. Granola, Fireflies, whatever you use. Pasted in or connected. The proposal gets written from what the client actually said.

  • Claude Design. Takes the rough draft and puts it in your brand. Charts included.

Claude reads the call, pulls the closest past proposals, writes a draft against the rules. I spend 10 minutes arguing with it. Then it goes to Design and comes back looking like we paid someone.

Takeaway: none of this is clever. It's a folder, a text file and a transcript. The hard part is writing down rules you've been keeping in your head.

STEP BY STEP
How it actually runs

  1. Record the call. Granola, Fireflies, any AI note taker. You want the transcript, not your memory of it.

  2. Set up Cowork. One project. A folder of past proposals, and a skill file with your rules in it.

  3. Generate. Paste the notes in. Claude reads the skill, finds the closest past proposals, writes the first draft.

  4. Argue with it. Ten minutes. Fix the pricing, tighten the scope, read what it flagged. Change the skill, not just the document, so it's right next time.

  5. Claude Design. Paste the draft in, pick your design system, export the PDF. Send.

Write docs 4x faster. Without hating every second.

Nobody became a developer to write documentation. But the docs still need to get written — PRDs, README updates, architecture decisions, onboarding guides.

Wispr Flow lets you talk through it instead. Speak naturally about what the code does, how it works, and why you built it that way. Flow formats everything into clean, professional text you can paste into Notion, Confluence, or GitHub.

Used by engineering teams at OpenAI, Vercel, and Clay. 89% of messages sent with zero edits. Works system-wide on Mac, Windows, and iPhone.

THE BIT THAT MATTERS
Speed was never the problem

A client I work with got their proposals from a couple of hours down to about 30 minutes with an off-the-shelf assistant. No build. Just practice.

Then it stopped mattering. At five proposals a week, 30 minutes is fine. What was actually hurting was the proposals going out with gaps in them.

A gap becomes scope creep. Scope creep becomes hours you never quoted for. That's margin, gone quietly, and nobody sees it until someone compares actual hours to proposed hours months later.

So the skill has one section that isn't optional. Assumptions and Exclusions.

In the demo call, the MD says near the end: "and if it could do the invoicing too, that'd save the bookkeeper a heap." Throwaway line. On a busy day that lands in scope, unpriced, and it is not a small job.

The system caught it, put it out of scope, and gave it a variation path. That's the whole point. Not faster. Right.

Takeaway: if you score this on minutes saved you'll call it a win and lose money. Score it on actual hours against proposed hours.

WHAT DIDN'T WORK
The first draft priced hosting wrong

Honest bit. The first pass put hosting at a number that had nothing to do with our actual AWS costs. Had to go back and tell it $300 a month. It only knows what's in the folder, and the folder didn't have that.

And in real life I'd spend longer on the skill before trusting it. Ten synthetic proposals is a demo. Your real ones, with your real pricing history, is where it gets accurate.

Takeaway: it's only as good as what you feed it. So feed it your outcomes too. Actual hours, what got scoped wrong, what the client pushed back on. That's what turns a generator into something that gets better.

Refer The Rollout, unlock the AI stack

  • Every share gets you closer to the tools I actually use to run this thing. Hit the share button below and you're in.

    • 1 referral → 3 months free of Notion Plus — the workspace I run everything in

    • 5 referrals → 75% off Xero for 6 months — cloud accounting that handles the boring stuff so you don't have to

    • 10 referrals → 3 months free of Wispr Flow — AI dictation that's 3x faster than typing (I write half my newsletter with it)

    Your unique share link is right below 👇

Until next week,
The Rollout

Recommended for you

View all
caret-right