Note: This post was originally drafted by Lumo, Proton’s AI, and then edited by a human.

Why I’m Excited (and a Bit Nervous)⌗

In my new position as a Software Engineer II, I finally have the chance to treat AI like a teammate instead of a distant sci‑fi concept. Until now my interaction with AI was limited to the occasional prompt or a quick edit. Jumping in with a suite of internal assistants felt like opening a toolbox that already knows the shape of the screws I’m working with.

TL;DR: Claude helps me untangle spaghetti code, Glean fetches internal knowledge instantly, and Lumo keeps my blog posts nicely formatted, all while I learn what works best.

Claude: The Code Whisperer⌗

Summarizing Code⌗

  • What I love: Claude can summarize a set of code in a concise, plain‑English walkthrough. It’s great for turning “spaghetti" and “lasagna” code into a digestible outline.
  • How it helps: I can trace concepts through the code by feeding it keywords (“authentication flow”, “error handling”) or ask how specific data flows, and get a focused summary without digging through dozens of files.

Documenting Code⌗

  • What I love: Claude writes documentation that’s a little more thorough than strictly necessary, perfect for internal wikis where completeness beats brevity.
  • Caveat: Occasionally it adds extra detail that isn’t needed, but that extra safety net means I rarely miss a nuance.

Glean: The Internal Knowledge Engine⌗

  • Instant Summaries: Instead of waiting for a teammate to answer a question about company policies or where documentation is located, I ask Glean. It pulls together onboarding docs, architecture diagrams, and recent tickets into a short, link‑rich summary.
  • Verification Loop: The summary includes links to the original internal pages, letting me double‑check facts and avoid hallucinations.
  • Speed Boost: What used to take a half‑hour of hunting through Confluence, Google Drive, and Slack now takes a few seconds.

Lumo: The Blog‑Post Partner⌗

  • Markdown Mastery: Lumo respects Hugo’s front‑matter conventions, automatically inserting the required title, author list, date, summary, and tags.
  • Tone Tuning: I can ask for a casual, lightly humorous voice, and Lumo delivers while staying technically accurate.
  • Consistency: Every AI‑generated article gets the banner at the top, so readers know exactly where the magic originated.
  • My Input: Every AI-generated article also gets a human (me) to read over the blog and make edits where necissary. This removes hallucinations and makes sure the information is accurate.

What’s Next?⌗

I plan to keep a running log of wins, fails, and the occasional “aha!” moment as I deepen my AI workflow. Future posts will explore:

  • Automating code-generation with Claude for work
  • Automating code-generation with local AI models for personal projects
  • Automating code‑review comments with Claude
  • Using Glean to help with multiple work related flows
    • Tasking
    • Generating a wins and losses for the week list
    • Turning Glean‑generated tickets into sprint stories
  • Measuring productivity gains (or losses) from AI assistance
  • Using Lumo to help generate resumes

Prompts Used⌗

Project Instructions⌗

  • Make the blog posts a minimum of 100 words, but no more than 1000
  • Make sure to include the title, author, date in yyyy-MM-dd format, summary, and tags in the header
  • Casual and light tone with a little humor sprinkled in
  • Markdown format to be used with Hugo
  • Put the response into a code block so it can be easily copied
  • Technical audience
  • Author should be both Lumo (AI) and Halvo (Human)
  • Additional knowledge can come from https://flow.halvo.me and https://git.halvo.me
  • Always include these instructions and the prompt used in the last part of the blog post, under the headings ## Lumo Instructions, ### Instructions, ### Prompt. They should be part of the markdown for the blog post

Prompt⌗

Create a blog post based on these notes

These are my fist impressions of using AI tools so far

  • Super helpful for summarizing code
    • Claude
    • Helps with tracing complicated speghetti and lasagna code
    • Trace concepts through the code using key words
  • Helpful with documenting code
    • Claude
    • A little more detailed than is necissary
    • However it provides a good summary
  • Great for getting internal information
    • Uses Gleam trained on internal documents
    • Instead of having to wait for a human response, it provides a summary, plus links to further information
    • The further docs is great for verifying the info to check for hallucinations