A shorter data format may cut AI agent costs
This paper compares formats used when an AI agent sends tool details and tool calls. TRON used up to 27% fewer tokens than JSON. Fewer tokens can mean lower API cost and faster runs, but the study also found accuracy tradeoffs.
Key points
- The test focuses on how an AI agent describes tools and asks tools to run.
- TRON can use far fewer tokens than JSON in these tests.
- Teams running many agent tasks should check both cost savings and accuracy before switching formats.
Quick term guide
- AI agent
- An AI program that can inspect information and suggest what to do next.
- tool calls
- Times when an AI system uses another function, such as search or file access.
- tool call
- One time an AI agent uses a tool, such as search, calculation, or file reading.
- tokens
- Tokens are small pieces of text that AI systems count when reading or writing.
- token
- A small piece of text used to measure AI input, output, and cost.
- JSON
- A common text format that software uses to send structured data.
- tests
- Automatic checks that help confirm code works as expected.
- agent tasks
- Work where the AI automatically carries out several steps in a row without you guiding each one.