Databricks has a great write up of best practices for reducing AI token costs.
1. Adopt open weight/lower cost models. Most tasks don’t need the best models.
2. Harnesses are a form of lock-in, adopt one which allows model flexibility.
3. Avoid hard per developer budgets. Prefer dashboards and spend gates.
https://www.databricks.com/blog/managing-ai-coding-costs-scale