What an agent team delivered—and what it cost

For a consulting project, a client needed a professional-course curriculum prepared for submission to a licensing authority. The work required research, original content, and attention to regulatory requirements.

I used a team of five AI agents in Claude Code to divide the work.

Give each agent a job

• Find: map the competitive landscape of approved course providers.
• Evaluate: examine competitor curricula and identify gaps.
• Evaluate regulations: read the regulatory code and licensing requirements.
• Build: write the course after the research was available.
• Check: audit the output for coverage, compliance, and originality.

Research ran in parallel. The builder waited for that work, and the checker reviewed the result. At one point, the checker caught that only the first chapter had been written and flagged it. The builder completed the remaining content.

What the workflow produced

The output contained 80,061 words across eight chapters, with lessons, glossaries, summaries, quizzes, a final exam, and regulatory citations.

The workflow’s internal checker assessed the material as ready for submission. That was an internal assessment, not approval from the licensing authority.

What it cost

The session exhausted my daily AI credit limit. I upgraded my plan to finish the work.

Agent teams can consume a lot of capacity because each agent maintains its own context. Parallel work speeds up parts of the process, but costs can rise quickly.

The lesson

The most useful part was building review into the workflow. Research, drafting, and checking had distinct responsibilities, and the checker caught an incomplete deliverable without a separate request from me.

That doesn’t make the approach appropriate for every task or eliminate the need for judgment. It does show how much the design of a workflow matters—and why cost belongs in the evaluation alongside speed and output quality.