Module 14 · Lesson 03
Rolling Out Claude to New Team Members
Reading time: 14 minutes Track: Claude Fluency for Teams · Lead/Manager path
The new member onboarding problem
New team members who learn Claude without guidance tend to develop one of two bad habits quickly:
- Over-trust: accept Claude outputs without verification, ship things they don't understand
- Under-use: have a few frustrating early experiences and conclude Claude isn't useful for real work
Both habits are sticky and hard to correct later. Getting the first week right prevents months of course-correction.
The first-week onboarding structure
Day 1: Setup and foundations (1 hour)
- Account setup with correct plan tier
- Data classification briefing (critical — before any substantive use)
- Introduction to team CLAUDE.md and shared conventions
- Where to find the team prompt library
- One hands-on exercise with a low-stakes task in their actual work
Keep Day 1 light on concepts and heavy on getting them actually using Claude safely.
Days 2-3: Guided practice (30 minutes/day)
Pair the new member with an experienced Claude user for their first real tasks. The goal isn't to watch — it's to help them develop good prompting habits from the start:
- How the experienced user structures prompts
- How they iterate when output isn't right
- How they review and verify output
This live modeling is worth more than any written documentation.
Days 4-5: Supported independence
The new member works independently but debrief daily (5-10 minutes): what worked, what didn't, what did they try? This feedback loop catches bad habits early and surfaces good patterns to add to the shared library.
Exercises that build the right habits
These exercises work well for onboarding because they combine immediate value with pattern-building:
The context window exercise: Ask Claude to help with a task, then try again with significantly more context. Compare the results. Internalizes the lesson that context drives quality.
The verification exercise: Ask Claude for a fact you can check. Verify it. This happens to reveal a hallucination often enough to make the verification habit visceral rather than theoretical.
The refinement exercise: Start with a vague prompt, get a so-so output, refine it specifically three times. Builds the habit of specific iteration rather than vague dissatisfaction.
The CLAUDE.md exercise: Walk through the team CLAUDE.md and ask: "What would have gone wrong if Claude didn't know this?" Builds understanding of why good project context matters.
What not to do
Don't start with agentic tasks. New users should develop judgment about Claude's outputs before giving it autonomy. Establish the "stay in the driver's seat" habit first.
Don't skip data classification. Every day without a data classification briefing is a day someone might paste something they shouldn't.
Don't just hand them a prompt library without context. A prompt library without understanding of why the prompts work leads to cargo-culting, not learning.
Don't set expectations of perfection. New members who expect Claude to always be right have worse outcomes than those who expect to iterate. Set accurate expectations: Claude is a powerful starting point, not a final answer machine.