---
id: con-onboard-ai-like-a-new-hire
schema: concept/1
layer: garden
status: emerging
name: Onboard AI like a new hire
aka: ["hire and onboard your AI", "AI copilot setup", "give the agent its own accounts", "earned access", "AI employee"]
summary: >
  Setting up an AI assistant or agent the way you'd bring on a new colleague: a role and ways of working,
  background on the organisation, its own limited accounts, and more access only as it proves reliable.
  Described by Tal Raviv for copilots and by Claire Vo for personal agents.
first_seen: "2025-07"
relations:
  builds_on: [con-custom-instructions, con-projects-and-knowledge]
  evolved_from: []
  related: [con-delegating-to-ai, con-custom-assistants, con-privacy-and-data, con-agents]
  parallels:
    - idea: "a probation period"
      field: "human resources"
      note: "A new hire gets a laptop and a few systems on day one, and wider access once their manager has seen their work."
implementations: []
competencies: [cmp-describe-the-task, cmp-steering, cmp-use-responsibly]
origin:
  kind: ai-proposed
  by: "seed agent 2026-10-16"
  rationale: "Two sources reached the same model independently, and it is one nontechnical learners already have from managing people."
---

Onboarding is about the assistant you'll work with for months, not one task. Two writers arrived at this picture separately. In July 2025, Tal Raviv laid out his setup for a work copilot on Lenny's Newsletter in steps he calls "hire" and "onboard": write how it should work with you, including when to push back, then give it background on your role, team and goals. In 2026, Claire Vo described giving each of her OpenClaw agents its own email address and calendar, as she would a new employee, and widening access only after watching how they behaved.

The new-hire picture also settles two questions people get stuck on. For what it should know, think of what a new colleague would need in their first week. For what it may touch, give it only what the job needs, under its own name, so you can see and undo what it did.

## Example

An events manager's new assistant can read the venue inbox and draft replies, which she sends herself. After a month with no bad drafts, she lets it answer routine questions about dates and prices.

## How it connects

- Uses **custom instructions** and **projects** for the role and the background.
- Own, limited accounts are a **privacy and data** decision.
- Each task you give it afterwards is an act of **delegating to AI**.

## Try it

Write a one-paragraph "first week" note for an AI assistant: its role, three things it must know about your work, and one thing it may not do yet.