I swear every “simple API integration” starts the same way.

“Just call the endpoint and parse the JSON.”

Then reality shows up:

timeout
random 502
429 because apparently 1 request every 30 seconds is “too much”
response schema changes
one missing field breaks everything
retry logic retries the one thing that absolutely should NOT be retried

And suddenly the 8-line script has become a tiny distributed systems lesson.

The part I hate most isn’t even the API failing.

It’s when it fails just often enough that you start doubting your own code.

Was it me?

Was it the network?

Was it the provider?

Was Mercury in retrograde?

At this point I trust boring code more than “clever” code:

explicit timeout, limited retries, backoff, logging, and please — PLEASE — tell me what actually failed.

The older I get, the more “works on my machine” feels like a threat.