Writing from the perspective of engineers building long-term systems.
Blocking GPTBot does not take you out of ChatGPT search. Blocking OAI-SearchBot does. How to tell AI crawlers apart, and why your CDN may be overruling your robots.txt.
A free AI feature is a meter running on your card. How we cap speech, tokens and voice per session and per day, and what the product does when it reaches a cap.
Practical patterns for voice AI turn-taking: speak sentence by sentence, let people interrupt, cache what repeats, and show captions while they talk.
Uptime is a lagging indicator. Resiliency is a leading indicator. How we architect for the inevitable crash.
The different time horizons required when building systems that sit at the bottom of the stack.
Why technical constraints usually mask product requirements, and how to rigorously interrogate the 'real-time' need.