Unverified50% confidenceEventExact time
In one "add 20 questions" operation, the author's Agent had the model iterate for 9 consecutive rounds, burning approximately 150,000 tokens before giving up and telling the user the API was broken.
1
Sources
50%
Confidence
Long-term
Relevance
8/31/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedNateHerk's team once mass-emailed between 150,000 and 200,000 people with an incorrect discount code because an AI agent misunderstood the task.56% similarUnverified模型自主触发Skills时，全量注入所有Skill描述的方案在Skills数量超过数十个后性能显著下降56% similarUnverifiedAgents suffer from error accumulation in multi-step tasks; a ten-step sequential process where each step has 95% accuracy results in an overall success rate that drops to approximately 60%.55% similarUnverified全栈优化的叠加效应使得过去18个月内每token的推理成本下降了约10倍54% similarUnverified在FastAPI开源项目（53个文件）上用20个问题基准测试，不使用工具每问83,000 tokens，使用本地索引工具每问4,900 tokens，减少94%54% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/827559API
curl https://kongchang.com/api/v1/knowledge/claims/827559MCP
get_claim(id=827559)