After this lesson, you will be able to:
- Name the standard failure classes that arrive with real traffic, and the hardening measure that answers each
- Sequence a hardening pass when the calendar will not fit all of it
- Add the LLM-specific production layer: guardrails, cost controls, and drift monitoring
- Sell hardening time to a customer who thinks the demo already works
Your prototype works. A real user would miss it if it vanished, the sponsor has seen it, and the pressure now is to roll it out to everyone, immediately, as is.
This is the most dangerous compliment an FDE receives. The distance between a prototype that works and a system that survives is real, invisible to everyone watching the demo, and entirely your job to defend.
#A day of traffic, simulated
The failures that find production systems are not exotic. They are a short, standard list, and each one has a known, affordable answer. Run the day below a few times: first with nothing enabled, then adding one measure at a time.
Notice what the simulation is claiming. Not that disasters might strike, but that these six arrive on an ordinary day: a rate-limited API, an oversized input, a curious colleague with a URL, a slow dependency, a provider outage, a security scan. None is rare. The only variable is whether each one is an incident or a log line.
The calendar allows two hardening measures before Monday's rollout, not six. Judging by blast radius, which two go first?