Writer’s brief

The hook

I had a call yesterday with a potential new client and by this morning I had a full proposal drafted, without touching a single app manually.

Main points to cover

  1. Granola records calls locally and caches transcripts - you can pull them programmatically even if the sync fails
  2. Claude Code can search Slack, find the relevant conversations about a deal, and synthesize context from multiple threads
  3. Going from raw transcript to structured proposal deliverables is a 15-minute workflow when the AI has access to your communication history

The angle

The “AI assistant” thing people talk about actually working in practice. Not hypothetical, not a demo. A real client call, a real proposal, real Slack threads.

Target reader

Freelancers, consultants, agency people who do a lot of calls and then spend hours writing up proposals. Anyone using Granola or similar meeting tools.

Tone notes

Casual, show-don’t-tell. Walk through what actually happened. The interesting part is how many systems talked to each other (Granola cache, Slack, file system, clipboard) without any of them being “integrated” in the traditional sense.

Raw material / moments to write from

  • Alex couldn’t remember the company name (“n2t2 or t2n2 i can’t remember”) but we found it anyway by searching
  • The transcript was tracked in the sync log but the actual file never made it to disk - had to go into the raw Granola cache
  • The Granola cache has nested JSON (JSON string inside JSON) which is a fun quirk
  • Speaker labels come through as “?” - Granola doesn’t reliably tag who’s talking in the raw data
  • The Slack search surfaced Britt’s original briefing notes, Alex’s call notes he’d already posted, the pricing negotiation thread, and the rate discussion - all from different timestamps and threads
  • Alex thought Britt said she COULD raise S&J prices but the actual Slack message said “I cant raise the prices on S&J” - the AI caught a detail the human misremembered
  • The proposal went through 3 iterations in the conversation: first version was too structured (entity-per-month), final version was looser and more realistic about how strategy work actually unfolds
  • Britt has an automated proposal builder (proposals.site) that takes Granola notes and generates proposals - the future is weird