Topic hub
AI Agents
Build autonomous AI agents that reason, use tools, and accomplish complex tasks.
All posts · 47
AI AGENTS01
Anthropic API Error 529 Overloaded: Agent Retry PolicyThe official SDKs retry a 529 twice, for 1.5 seconds of nominal backoff. Here is the branch table an agent loop needs, including three 429s.
AI AGENTS02
The 400 names input_json_delta because it is a streaming delta type, never a request-side content block. Six documented breaks in streamed tool_use assembly.
AI AGENTS03
Is It Safe to Open an Untrusted Repo? 12 Coding Agent DefectsTwelve fixed defects across three vendors let a repository configure the coding agent before the trust prompt. What trust gates, and what it never does.
AI AGENTS04
tool_use ids were found without tool_result blocks: 6 FixesThe 400 does not always mean a result is missing. In two filed reports every result was present, just not ahead of the other blocks in that message.
AI AGENTS05
vLLM Prefix Cache Not Hitting on Multi-Turn Agent RequestsvLLM measured a 1.7 percent prefix cache hit rate on 610 agentic traces. The usual cause is your chat template, not your cache size.
AI AGENTS06
AI Agent Says Done But Did Nothing: Detect False SuccessRoughly 45 to 48 percent of agent failures close with a confident completion claim. LLM judges catch almost none of them. Here is what actually does.
AI AGENTS07
AI Due Diligence Data Room Review: From Index to Request ListPreregistered testing put hallucination at 17 to 33 percent in retrieval-backed legal AI. Here is the four-pass data room review that survives a dispute.
AI AGENTS08
AI for AML Alert Triage: Cutting a 95% False Positive RateRule-based AML monitoring runs 90 to 95% false positives at 30 minutes an alert. Here is the triage pipeline, the evidence contract, and the validation path.
AI AGENTS09
MCP Error -32000: Connection Closed (6 Causes and Fixes)MCP error -32000 is not a timeout and not an MCP protocol code. One public debug log fails at 6544ms inside a 30,000ms budget. Six causes, in triage order.