Notes from things that went wrong first
Written by the engineers who were there, mostly about decisions that turned out to matter more than anyone expected at the time.
Ask for the peak message rate, not the average
The single question that changes an IoT architecture more than any other, and why the answer is almost never the number you are first given.
Signed OTA belongs in the first prototype
Over-the-air update is treated as a launch feature and built last. That ordering is how fleets end up stranded on firmware nobody dares replace.
Why your operators ignore the dashboard
Most operator consoles are built for the demo rather than the shift. The gap between those two audiences explains almost every abandoned dashboard.
Your predictive maintenance problem is a labelling problem
Teams arrive with years of maintenance history and ask which model to use. Almost always, the answer is that the model is not the bottleneck.
Disagree with any of this?
Several of these positions came from being wrong in public first. If your experience says otherwise we would genuinely like to hear it.