We haven't really talked about token budgeting in the workshop materials so far (I should add it to the intro), but here is an opportunity. A common recommendation is to use a stronger, more expensive models for planning and cheaper models for the implementation. The idea is the more expensive model can do the more complex work of analyzing your code base and making architectural decisions; the cheap model just needs to be good enough to follow the plan and handle unanticipated errors that come up. If the agent is not able to stick to the plan, then it may be necessary to revisit the plan or switch back to the heavier model for implementation.
We haven't really talked about token budgeting in the workshop materials so far (I should add it to the intro), but here is an opportunity. A common recommendation is to use a stronger, more expensive models for planning and cheaper models for the implementation. The idea is the more expensive model can do the more complex work of analyzing your code base and making architectural decisions; the cheap model just needs to be good enough to follow the plan and handle unanticipated errors that come up. If the agent is not able to stick to the plan, then it may be necessary to revisit the plan or switch back to the heavier model for implementation.