AI Agents · Course resources
Notes: Measure AI Agent Costs and the Work You Still Do
Review the key ideas from Measure AI Agent Costs and the Work You Still Do.
Section 9 measures the work around an agent-assisted job, how that work reaches people, its complete cost, what can be reused, and how to propose a bounded pilot whose result can later be judged.
| Lecture | Understanding developed |
|---|---|
| What Work Remains After AI Automation? (Preparation and Exceptions) | Preparation, review, exceptions and maintenance remain part of the complete job. Coverage by item count does not establish working time saved. |
| How Do You Manage Work Left by AI? (Review Queues and the Last Mile) | A review queue can reduce interruptions, while urgent work still needs a route that preserves its deadline and evidence. |
| What Does It Cost to Run an AI Agent? (Tokens and Usage) | Complete cost separates allocated plan cost, usage and services, human time, setup, maintenance, retries and failures. Unmeasured values remain visible. |
| What Makes an AI Request Reusable? (Prompt Templates) | A prompt template keeps purpose, criteria, output and stopping boundaries stable while marking every input that must change for a later job. |
| How Do You Propose an AI Agent at Work? (Business Case and Pilot) | A pilot proposal sets the job, duration, inputs, quality standard, retained human work, exception ownership, complete costs and continuation evidence before a trial begins. |
The Amazon Vulcan statement remains tied to item types. The complete cost example remains illustrative, and Priya's pilot remains a proposal with no observed result. Worksheet labels come from the target source file; publication state is not inferred.