- Issue created by @marcus_johansson
One of the strength with the AI Agents 1.0.0 pattern, was that it was consistent when tools ran and with an consistent amount of tokens used.
If you had an agent that was made for creating a taxonomy term, you knew which tool should be run - first validation that it doesn't exist, validation that the vocabulary exists etc and then creation.
Many AI providers has a "force_tool_use" where you can say that one tool has to be used. This means that in the above case we could define that validation tools runs on the 1 run and creation on the 2 run, to make sure that its more stable.
That would remove the last "advantage" the old system had.
On the detailed tool usage for any tool, we add a free text field that say something like "Force use on nth run"
In this tool you can give back one or many numbers.
When the agents is running the tools, on that nth run check if there are any tools that should be forced and set them to forced.
Active
1.1
Code