Minnesota Loons & AI Fashion: Agrima's Weekly Work Log

Most work updates turn into highlight reels: the wins get polished up front and the messy middle gets left out. This video takes the opposite approach. Humanitarians AI Fellow Agrima walks through an actual, thread-by-thread log of one week's work, built with Claude, and shows exactly how she went from a raw data dump to something readable without inflating or hiding anything.

Skipping the highlight reel

The week split two ways: writing about AI and hands-on field work on the LON project, which tracks Minnesota's common loons by drone. Rather than summarizing the week after the fact in the most flattering light, Agrima set out to log it as it actually happened, item by item, with real statuses attached to each piece of work.

Coding the status script in Claude

The build started simple. Agrima asked Claude for a script that logs what happened during the week, thread by thread, using one dictionary per thread and a log function that prints each item with its actual status. There was no scoring and no tally, just a record of what happened. Running it produced two published articles and six items on the LON project, three done and three still in progress. It was honest, but it read like a data dump, not something anyone would actually want to read.

Refining the output into a readable log

The next step was asking Claude for a real fix: turn the data dump into something that reads like an actual log, with a standout line at the top of each thread calling out the week's headline before every entry underneath it, still in order, still with every real status. Running the updated script produced the same two threads and the same eight real items, now organized with a standout summary up top for each. Nothing was hidden and nothing was inflated, just an actual log of the work.

Thread one: AI in fashion forecasting

On the writing side, the week produced two articles: one on how AI is reshaping fashion forecasting and design, and one on the open-source model gap closing. Both count as real, logged output for the week, not a vague summary of "wrote some things."

Thread two: the LON drone project

On the field side, the LON project tracks Minnesota's loons by drone, and the week's log shows real ground covered: a drone now in hand, the FAA certification path researched, and a strategy written for reaching lake communities and boaters. Certification, footage collection, and a trained model are still in progress, and the log says so plainly rather than glossing over what isn't finished yet.

Key takeaways

• A highlight reel hides the messy middle; a real log records every item with its actual status.
• Claude can build a simple logging script from a plain description: one dict per thread, one log function, no scoring.
• The first draft of a status script often reads like a data dump; asking for a standout summary line per thread makes it readable without changing the underlying data.
• Splitting work into threads, writing and the LON drone project in this case, keeps a log organized without losing any items.
• Being honest about what's still in progress (certification, footage, a trained model) is part of what makes the log trustworthy.

Who this is for

This is a practical example for anyone tracking their own project work, especially other Humanitarians AI Fellows, who wants a faster, more honest way to report progress than a polished recap. Try it yourself: list what you actually worked on this week, in order, and ask Claude to turn it into a log, not a highlight reel, with a standout per thread.