OneCrewAIOneCrewAI
    Scope the Work

    Ep 10 · Level 2 · Chatbot, workflow or agent

    This video was made by an agent

    The season's payoff: every OC episode is made with all three. A chat, a workflow, and an agent on a short leash.

    Watch · 1:37

    This video was made by an agent, and so were the others. Here's how, using everything from this season.

    1 · A chat: thinking it through

    Every episode starts as a conversation: the topic, then the script, line by line. A person decides what goes in, like swapping Zapier for n8n in the workflow episode.

    2 · A workflow: same steps, every episode

    Record the voice (two takes), time every word, draw every frame, render the video. Predictable, and every run can be checked.

    3 · An agent: running it all

    Claude Code, from Anthropic, gets the goal: make episode ten. It writes the scenes, records, renders, checks its own stills and fixes what's off.

    The dial: draft and ask

    It stops at every checkpoint: the script, the voice take, the storyboard. Nothing renders until a person says go.

    The keys: it can't publish

    It works on its own copy of the site. It can't publish it, and it can't post the video. A person does that.

    When something looked wrong (a hand, a wave, an arrow), the fix became a rule in its playbook. Every edge case becomes a step.

    Try it

    Your turn: which one will you try first, a chatbot, a workflow or an agent?

    Share your answer with OC in the comments@onecrew.ai

    Pick the simplest one that does the job. That's Level 2.

    Read the transcript

    This video was made by an agent. So were the others. Let me show you how, with everything from this season. Let's go.

    First, a chat. Every episode starts as a conversation: the topic, then the script, line by line. A person decides what goes in. That's the chatbot job: thinking it through.

    Then, a workflow. Same steps, every episode. Record the voice, two takes. Time every word. Draw every frame. Render the video. Predictable, and every run can be checked.

    And running it all, an agent. Claude Code, from Anthropic. It gets a goal: make episode ten. It writes the scenes, records, renders, checks its own stills, and fixes what's off.

    But look at the dial. Draft and ask. It stops at every checkpoint: the script, the voice take, the storyboard. Nothing renders until a person says go.

    And the keys. It works on its own copy of the site. It can't publish it. It can't post this video. A person did that.

    When something looked wrong, a hand, a wave, an arrow, the fix became a rule in its playbook. Every edge case becomes a step.

    That's the season. A chatbot to think it through. A workflow for the same steps, every time. An agent when the steps keep changing, on a short leash. Pick the simplest one that does the job.

    Your turn: which one will you try first? Tell me in the comments.

    I'm OC. Follow so you don't miss what's next.

    Bring it to your team

    These lessons are the start of our AI Fluency workshop: four hours, your team, real work.

    See the workshop