Keo Sar

The newsletter on Monday highlighted a launch that centered on agent teams that could be opened up to find out where things had gone wrong. This sentence has stayed with me all week.

Most of the demonstrations with agents focus on the final output. But the part that interests me is the failures. If an agent chain produces a wrong answer, then I have no clue where along the way something has gone wrong: was it during the retrieval, the function call, or an incorrect assumption made three hops back?

I created a simple invoice management system called Tallyleaf in a single weekend using Cursor and Claude Code. It worked flawlessly on my laptop, which should've been enough to raise a red flag.

When the first time came to edit a record simultaneously by both my friend and I, one of us simply overwrote the other's work without any warnings or errors. Agent does not take concurrency into account since I didn't include it in my prompt, and thus it doesn't create anything that wasn't explicitly asked for.

There was an option that defaulted to "log everything" in staging and "log nothing" in production, which seemed to make sense until we discovered that it had never been explicitly set for the new service. The new service would just inherit the logging behavior from production, which is none at all. We didn t see a thing going wrong for three weeks straight in one of our services.

The issue was not in the logic. The issue was that the reasonable default value meant there was an implicit decision made that nobody decided on.