Building Klau in the Open, Week 11: 60 Commits, No New Features

Week 11 of building Klau in the open.

60 commits. Most of the work this week was invisible to anyone but me.

An enterprise pilot is doing what pilots are supposed to do: real operations, real data, real edge cases. My customer runs container sizes that Klau had never seen, which led me to hardcoded assumptions in eight different places. They also have users in multiple time zones, which needed some UI cleanup. These things never show up in a demo environment. Every one found is another one fixed.

You don't know what you don't know until a real customer shows you. I love this part.

I spent a lot of this week consolidating the job creation logic. Every path that creates a job did it differently: the dispatch board, imports, the ordering bot, and the integration layer. Different code, same outcome, and sure to drift apart over time. That made sense during the rapid feature build-out, but it was time to clean up, because that's where bugs live. One shared service now handles all of it. It's cleaner, it's testable, and I can trust it as real users put Klau to the test.

I also caught a runaway AI spend problem. A background job was calling the API every night on a schedule nobody needed, quietly using up tokens. I found it, removed it, and added cost tracking so it can't hide again.

This is the part of building software nobody posts about. No new features this week, and fewer lines of code at the end than at the start. Just making the thing more reliable.

You can see Klau on the storefront.

Originally posted on LinkedIn on March 20, 2026.