Laura Rosenberger

Category: Field Notes

  • Voice for Email

    I don’t like the idea of AI connecting directly to my Gmail. My entire life is in there.

    I don’t want it messing with my tone of voice or making people think I’m using fake AI responses. Often there’s not enough context for AI to write accurate emails anyway.

    But there is another way I use AI to save time in my inbox.

    The Solution

    I use a voice-to-text app with a custom AI prompt to format emails as I like them.

    Tools like Monologue, WhisperFlow, or SuperWhisper let you speak your email reply. The prompt structures it the way I like, removes filler words, and cleans up stumbles. It doesn’t transcribe word-for-word – it turns spoken thoughts into clean, natural-sounding emails.

    And I don’t have to give access to my inbox to any AI tool.

    How I Use It

    I use Monologue (https://www.monologue.to/). When I need to reply to an email, I hit the option key twice on my keyboard and just talk as if I’m on the phone. The AI processes the audio, cleans it up, and pastes it straight into Gmail for me to review and send.

    The prompt handles the polish whilst keeping my voice. No risk of AI accidentally sending something. Full control over what gets sent.

    Why This Has Impact

    I fly through my inbox now. The barrier to replying is much lower.

    It’s not just that speaking is faster than typing. It’s that replying to an email suddenly feels like talking to someone in the room instead of sitting down to “write an email.” That psychological shift has had the biggest difference.

    The friction disappears and the risky automation stays out.

  • What is left once Claude Code can write the code

    More and more people are looking at Claude Code for non-technical use cases. Claude Code handles the code, but what does a human still need to do?

    A fractional COO friend has been watching some Claude Code webinars and is ready to start playing with it.

    One of his first projects is to use it for invoice generation. We spent a few minutes on whether Claude Code could do this. Yep, it certain could code something. But what else does it need to do this well?

    Accurate timesheet data. Up-to-date client information and rates. Expense policies. Receipts. Knowing where the data lives and whether there are processes behind it so that you can trust it to be correct and up to date.

    All of this was his bread and butter as an operator, and it's easy to forget this doesn't come easy to everyone.

    Once Claude Code handles the coding, what remains is right in an ops person's wheelhouse.

  • Software Engineering Tutor

    I was blindly accepting everything Claude Code suggested. Commands would run, code would get written, and I’d just click “yes” without understanding what was happening. I wanted to actually learn what I was accepting.

    The Solution

    I created a project in the Claude desktop app to act as my software engineering tutor. The prompt is simple:

    You are my software engineering tutor. Explain code, tools, and technical concepts in plain language, breaking down unfamiliar ideas step by step. Check my understanding with questions.

    The key part is “check my understanding with questions.” This forces me to actually verify I understand what I’m learning.

    How I Use It

    I keep two windows open:

    1. Claude Code in Terminal – building the actual project
    2. Claude Desktop App – Project: Software Engineering Tutor – my learning companion

    When Claude Code does something I don’t understand, I copy the code or command into the tutor project and ask for an explanation. The tutor breaks it down step by step, then asks me questions to verify my understanding.

    That’s when I realise I didn’t understand as much as I thought.

    Why This Has Impact

    I’m learning instead of just blindly clicking “accept”. I don’t need to write production-grade code, but I want to be able to read a bash command and know whether it’s about to delete my files or create a backup.

    The tutor approach works because it’s active learning. Having something explained and then being tested on it? That actually sticks and I’m learning 10x faster.