LLM
Your AI App Slows at Peak? It's Not the Model — It's an Untouched Config Line
An AI writing tool with 50K DAU saw slowest users jump from 1.2s to 18s at evening peak — culprit was an untouched default config line, not the model.
Back to home
Engineering Practice
4 articles tagged with this topic
Agent
Agent Projects Don't Fail Because Models Are Dumb — They Fail After HTTP 200
An engineer recaps an Agent failure chain on Juejin: errors collapsed into one code, but real crashes scatter across 7–8 stages, mostly after HTTP 200
Agent
AI Agents: From Demo to Production, Streaming Is the 'Wait' Divide
For AI agents, the divide is the wait: streaming cuts first-token latency from seconds to hundreds of milliseconds, making or breaking retention.
AI Assistants
After AI Assistants Launch: 90% of Failures Aren't in Algorithms
Beyond AI Agent hype, engineers' checklists reveal: parameter validation, retries, and context caching are the real deployment threshold.