The lab.
We roast coffee in San Jose. We also spend a lot of time figuring out how to do things better with two people and a small roaster.
Some of that involves ChatGPT and Claude. We use them to work through roast questions, build the website, and solve some of the less glamorous problems of running a coffee business.
At the roaster.
I use Claude as a sounding board when I'm reading roast notes, weighing different approaches, or deciding what to try next. Some suggestions are useful. Others don't hold up.
Either way, we still have to roast the coffee, brew it, and taste it.
Jean-Marc
Building Folio.
We built this website with Claude Cowork. When we have a new coffee to release or want to change something, we can do it ourselves. It's been fun figuring out what we can make.
The Reserve Manager.
Running a coffee membership turned out to be trickier than we expected. Just because someone pays this month doesn't mean they're owed this month's coffee.
We've been working through that problem with ChatGPT, figuring out how to track members, monthly issues, the coffee they're owed, and what we've already sent.
Now we're building a small tool with Lovable and Supabase to keep it all straight. It's still a work in progress.
Five batches, one hot afternoon.
Nº 06 is a washed coffee from Duromina, in western Ethiopia. We roasted a 500g test batch first to learn it, then four more at 660g. Every batch did the same thing: about 30 to 45 seconds into first crack, the roast slowed way down.
Jean-Marc tried four different ways of handling the heat at crack. The dip got shallower or deeper. It never went away. We cupped all five the next day and settled on the light end, City to City+.
Claude can read our notes. It can't smell first crack, taste a cup, or tell us whether a coffee is worth sharing.
This is where we'll share what we're trying, including the things that don't work.
See this month's coffeeQuestions, or something you'd like us to try? hello@folio.coffee