The CPCT Analysis

CPCT (Cost Per Completed Task) is a modelled agentic workflow that emerged spontaneously from the question, how do you model the cost of running an agent?

The depth behind that research is outlined on the Notebook post - Agentic Modelling - What Does it Cost to Run an Agent?. Most agents that would run in a business or enterprise setting rely on API tokens; therefore, the cost to run an agent is variable and can be extremely volatile depending on the task.

The method itself is developed out in the open, so it is far from completed, but the gist of it today is:

• start with a core workflow that has the potential to become agentic
• model it out as a detailed workflow, step-by-step, and add in the proposed agentic layer.
• run it through the CPCT across multiple turns and tasks, trace the agents and their associated efficacy, cost, etc. on each task
• capture the aggregate data and model it out as a workflow on a P&L

The result is a succinct, actionable analysis you can make a decision with, and it's yours. At this stage, I don't implement, resell tools, or take referral fees, so the recommendation has nothing to sell you except its own reasoning. Fair warning, offered as a feature: the method is young and improving in public. You're buying the current best version of it, and your workflow makes it better. Updates to the latest iterations and research on the CPCT will be published on the Notebook.