Four patterns for conversations that stay productive. When things go sideways, the problem is almost never the tool. It is the structure of the conversation.
What did the adversarial sequence surface? What happened when you tried a different model on a real task? Open with wins and surprises from Module 3 homework.
Anyone who got surprised by what the skeptic pass found has the opening story. The adversarial sequence was designed to surface things you would not have caught on your own. What did it catch?
"Who has had a conversation go sideways in the last few weeks?" No lecture. The room surfaces the pain that the patterns solve. Lost context, diminishing returns, conversations that started strong and drifted.
This is the first session of Phase 2. Phase 1 gave you the foundation: how AI works, how to give it context, how to push back. Phase 2 is about fluency. The goal across the next three modules is for you to become a confident, independent operator who explores actively and makes deliberate choices.
Participants describe what they have figured out in their own Projects. How they structured their work. What standing instructions they set. What they learned from sustained use over the past three weeks.
The four patterns get introduced as vocabulary for what participants are already doing, not as rules imposed from outside. The assumption disclosure instruction, the meta-learning prompt, the markdown handoff: these are participant-invented techniques that now get names.
Introduce each pattern with examples the room recognizes from their own work. The Setup Frame structures the opening. The Iteration Loop makes multi-turn deliberate. The Branch Strategy answers "should I keep going or start fresh?" The Summary Checkpoint preserves progress when you do start over.
Each pattern is explained with before-and-after examples. The full reference is on the Conversation Architecture page.
Anthropic's AI Fluency Index found that only 30% of users set explicit collaboration terms before starting a conversation. Users who did showed significantly higher fluency across every other measured behavior. The Setup Frame is the answer to that gap.
Each participant applies one pattern to a real piece of work from their Claude Project. Pick the pattern that matches a problem you are actually facing. Try the Setup Frame on a task you have been meaning to start, or run a Summary Checkpoint on a conversation that has gone stale.
At the end, each person names which pattern they used and what it changed.
Brief walkthrough of how Claude's memory and conversation search features complement the manual patterns you just learned. The platform is building infrastructure for what you are doing by hand. Both matter: the manual discipline builds the instincts you need to evaluate whether the automated version is getting it right.
Homework assigned. Two assignments, both practical, both tied to your actual work. Details on the homework page.
Next week we look at what happens when AI produces something that looks finished. Anthropic has some uncomfortable data on this. We will also step outside of Claude and explore the full model landscape honestly: what else is out there, how it compares, and when to use which tool.