5 posts
Training makes the headlines. Inference is what actually shows up on the bill, every request, forever, and it now eats over 80% of AI GPU spend.
No single model satisfies every workload. Real companies now route requests across fast, reasoning, vision, and local models instead of picking just one.
'Application plus LLM' was never the architecture, it was the demo. Here's the eight-layer stack a real AI system actually needs, and a map to every layer.
A 200 OK and a wrong answer. Traditional logs can't tell you why an agent decided what it did. Here's what to actually record, and how to trace it back.
The model call is the easy 5%. Here's the architecture, security, cost, monitoring, and reliability work that turns a demo into a system you can trust.
