New-hire onboarding is one of the most underrated AI use cases. Done right, it compresses time-to-productivity from 60–90 days down to ~30. Done wrong (or not done at all), every new hire reinvents the same questions you've answered for the last 8 people. Here's the workflow.
Create a dedicated Project per role ("Sales Onboarding," "Engineering Onboarding," etc). Load:
— Role-specific SOPs and process docs
— Anonymized examples of "good work" in this role (proposals, code review samples, customer responses)
— Common questions previous hires have asked in their first 30 days
— Your company values, voice guide, and any "this is how we operate" docs
— Org chart with who-owns-what
System prompt: "You are an onboarding assistant for new [ROLE] hires. Answer questions in the voice of the company. Cite specific docs in the knowledge base. If a question is outside what's in the knowledge base, say so — don't make up answers."
Most companies hand a new hire a PDF of policies and hope. AI-native onboarding hands them a Claude Project they can query.
Day-one orientation includes: "Here's the Onboarding Project. Ask it anything. \"How do we usually handle [scenario]?\" \"What's our standard process for X?\" \"Who would I go to about Y?\" It has answers."
Hires get unblocked at 11pm without bothering anyone. Their first-week question volume to coworkers drops 70%.
Give the hire a structured first-2-weeks plan with specific Claude prompts to run each day. Examples:
Day 1: Ask "Summarize what this company does and who we sell to." Day 2: Ask "What's the ICP and why does it look like that?" Day 3: Ask "Walk me through our standard sales process stage by stage." Day 5: Ask "What are the 5 most common objections we get and how do we typically handle them?" Day 8: Ask "Show me 3 examples of proposals we've won and tell me what made them work."
Each prompt teaches both the answer and the pattern of using AI on the job.
Tell the new hire: "This Project is for institutional knowledge and process questions. For anything time-sensitive, current customer issues, or financial specifics, talk to your manager. The AI knows what we've written down — not what happened yesterday."
Without this calibration, new hires either over-trust the AI (leading to wrong actions on outdated information) or under-trust it (defeating the purpose).
Every quarter, audit the Onboarding Project. What questions did new hires ask that the Project couldn't answer? Add docs for those. What's changed in the company that's now wrong in the knowledge base? Update it.
Quarterly maintenance: ~2 hours. Saves: 4–6 hours per new hire of redundant 1:1 questions.
1. Loading outdated docs. A Project trained on 2024 SOPs will give 2024 answers. Audit and refresh the knowledge base when major processes change.
2. Using a generic Claude Project for all roles. Sales onboarding and engineering onboarding need different docs and different system prompts. One Project per role.
3. Treating it as one-and-done. The Project decays without quarterly maintenance. Set a recurring calendar event.
4. Replacing human onboarding entirely. The Project supplements 1:1 onboarding — it doesn't replace it. Manager check-ins still matter. The Project reduces friction; it doesn't eliminate the human connection new hires need.
For typical knowledge-work roles, 15–30 hours per new hire in their first 90 days. Most of the savings come from eliminating redundant Q&A with coworkers.
Claude Team or Claude Enterprise. The shared Projects feature is the entire point. Without shared Projects, every hire builds their own personal knowledge base.
Don't put customer-specific data or financial details in the Onboarding Project. Use it for institutional knowledge (how we work, what we sell, who we are). Sensitive info belongs in restricted-access tools.
Partially — for technical/process knowledge, yes. For things like culture, relationships, and informal norms, you still need humans. Treat the Project as a 24/7 reference, not a replacement for managers.