02/25/2026
Something Alex kept coming back to during our WAF session last week:
AI is not this magic thing sitting off to the side that you don't know how to manage.
Every WAF pillar has AI implications that most teams aren't thinking about yet:
Reliability: What's your fallback when OpenAI rate limits hit? When a region goes down? When a retrained model starts giving different answers?
Security: Prompt injection is real. Users are typing English into your app and controlling its behavior. Content filtering isn't optional.
Cost: Token billing is a different world. Output tokens cost more than input. Every conversation passes the full history back to the model. A long system prompt adds up on every request.
Performance: Can you cache AI responses? Yes, with semantic caching. But the 5% difference in prompt similarity could change the whole answer.
Operations: Version your prompts. A/B test them. Watch for quality drift over time.
If you're adding AI to your Azure environment and not running it through the WAF pillars, you're taking on debt you'll have to deal with later.