- What changed
- A tested model chose direct tool calls over writing executable Python workflows in all ten tasks when both options were available.
- Why you should care
- Models may fail to choose executable workflows when direct tools are available, affecting agent design.
- Your move
- Test. Verify if models can learn to select executable workflows over direct tool calls.
- What to watch next
- Future benchmarks showing if models can learn to select executable workflows when direct tools are available.
- Event
- research
- Event date
- Sep 24, 2026
- Relevant to
- General AI readers