JUL 2026 · FIELD NOTES
An ongoing log of the design and engineering decisions behind Carom.
Enterprise software has a credibility problem. It’s a pain to buy—or even to figure out a price—and relies too often on slick salespeople selling to buyers who aren’t users. With a few notable exceptions, the “try a demo” link on a B2B site lands you on a salesperson’s call list, not in a working version of the software you want to buy. For companies selling to small and medium-sized businesses, there are fewer and fewer excuses for hiding their products from prospects—and it’s no longer fair to blame the difficulty of creating a good demo for the lack of one.
Software has a natural tendency to drift and sprawl, starting with a pristine design but ending up with a hodgepodge of similar but distinct ways of displaying data or accomplishing tasks. Avoiding that drift requires deliberate effort, and fixing it after the fact is hard. Fortunately, an LLM used right is a great tool to reduce drift.
There are two kinds of software. One demands that you feed it—your time, your typing, your discipline—before it gives anything back. The other reads the work you’re already doing, and gives back without asking. Call them extractive and ambient. Once you see the split, you see it everywhere. And you notice that every time we’ve been offered the choice, we’ve chosen ambient.
Like anyone in the software world—hell, the world—we’re big fans of coding with large language models. It’s amazing! Give Claude Code a few sentences and plenty of tokens, keep your laptop open, take a walk, and come back to a production-ready application made bespoke for you.
We like to sweat the details of every screen in Carom, especially the ones that communicate the most critical information. That means we ask a lot of questions, reject a lot of ideas, and refine a lot of concepts. Here are some of the decisions that led to the feature you'll see in Carom today.