Conversationalist

Multi-turn dialogue with context

Characteristics

Approach

Multi-turn dialogue with context

Turns per task

5-10

Mental model

AI as junior developer to guide

Error response

Provide more context, ask clarifying questions

Impact

Requests per feature

5-10 requests per feature

Rework rate

Moderate rework rate (30-40%)

Context quality

Context builds well but often diverges

What It Looks Like

User: "I need to add email validation to our user registration. 
       We use TypeScript, Jest for testing, and we have a UserService 
       that handles registration. We should check format and block 
       disposable email domains."
AI: "I'll create the validation with tests..."
User: "Looks good, but can you also add logging?"
AI: *adds logging*
User: "Great, now integrate with UserService"
AI: *integrates*
User: "Tests are failing on CI"
AI: "Let me check the CI config..."

Common Patterns

Good initial context
Still steering throughout execution
Reacts to issues as they arise
Some skill usage, but inconsistent
Better than L1, but still inefficient

How to Level Up

The shift

Stop steering and start delegating.

Practice

1.Start with: "Use writing-plans to create a plan before implementing"
2.Review the plan, approve it
3.Let it execute without interruption
4.Only intervene at checkpoints

Skills to learn

writing-plans — structured planningexecuting-plans — checkpoint-based executionverification-before-completion — verify before done

Measure success

Did your turns drop from 10 to 5?

Previous: Prompter Next: Delegator