Unverified60% confidenceOpinionExact time
In a typical AI feature deployment, the model calls themselves may account for only 5%–10% of the entire system's codebase, with the remaining 90% or more being engineering infrastructure built around the model.
2
Sources
60%
Confidence
Long-term
Relevance
8/9/2026
First Seen
Sources
AI落地的真正瓶颈不是智能：数据、工程与可靠性才是关键
Related Claims
UnverifiedIn AI application development, data preparation constitutes the majority of the workload, while writing model invocation code often accounts for only a small portion.79% similarUnverifiedAI tool adoption in the code-writing stage has exceeded 70%, but AI assistance rates in requirements review, architecture design, code review, integration testing, and similar stages are below 20%.76% similarUnverifiedA typical tiered deployment approach uses small open-source models to handle more than 90% of routine tasks, routing only complex multi-step reasoning tasks to frontier models, which can reduce overall AI operating costs by several times.76% similarVerifiedAccording to an AI engineer's account, 20% of their daily work is AI and the remaining 80% is pure software engineering.74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/717076API
curl https://kongchang.com/api/v1/knowledge/claims/717076MCP
get_claim(id=717076)