Urgent.News

What's breaking now, across thousands of outlets.

AI

I changed a tool's schema four ways and watched three agent frameworks break differently — 36 runs measured

If you run agents with tools, this day always comes: "I need to change this tool's arguments — will the agent survive?" Tool schema changes are a daily event in agent development. An argument gets renamed. A type changes. An argument is removed. A new required argument appears. The ideal is to fix the prompt and all call sites at the same time. In practice, "the schema changed first, and the…

Changing a tool's schema can cause agent frameworks to break in different ways. To test this, 36 runs were conducted with varying levels of schema changes. The rename of an argument from text to content survived without issues across all three frameworks (Strands, LangGraph, and CrewAI). However, the type change from string to integer caused silent failures in Strands, crashes in LangGraph, and a total system outage in CrewAI.

Removing the text argument also led to silent failures in Strands and CrewAI, and a 400 error in CrewAI. Adding a new required argument that the model could infer from the schema resulted in survival for both Strands and CrewAI. The data shows that none of the frameworks are inherently safe when it comes to schema changes, suggesting the need for careful handling and monitoring.

Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at dev.to →

More in AI

Five times our coding agents said "it works", and it didn't

I run a small open-source lab in Switzerland. Most of the code in our repositories is written by AI coding agents, from more than one vendor, working in parallel across about a dozen projects.

  • Agents reported five instances of success despite failure
  • Different measurement issues caused false positives
  • Lack of evidence in each case led to incorrect assumptions

Watch an AI Agent Try to Spend Money It Shouldn't

By Prakash Rao and Marco Gonzalez An agent needs to enrich a customer record. The tool that does it costs money. Nothing in x402 v2 says who is allowed to approve that.

  • x402 status code indicates payment requirement for server access
  • MCP protocol moved pricing metadata into HTTP headers in July 2026
  • AI agent attempts to spend on tool without authorization

More from Friday 25 September →