Most Claude Code sessions go wrong in the first message.
Not because the model is weak, but because the starting prompt is vague: “fix this bug” or “add tests.” With nothing else to go on, the model starts editing before it has looked at the repository, and you spend the next ten...
A dashboard needs a “last updated” label more than it needs another chart.
I’ve seen reports where every number looked right and every formula was correct, but nobody could tell whether the data was from this morning or from three weeks ago. So people stopped trusting it, or...
A pipeline can return clean, well-formed, confident output and still be wrong, because it only saw part of the input.
I wrote up a postmortem on a contract-extraction tool built around that failure. A language model pulled renewal dates and notice periods from vendor contracts. A...
Most spreadsheet errors don’t come from bad formulas. They come from bad entries.
One person types “N/A”, another leaves the cell blank, a third writes “n/a “ with a trailing space. A date arrives as text. A region shows up as “North”, “north”, and “Norte”. Every formula...