# An AI Chief of Staff for Senior Clinicians (2026): Decisions, Delegation, Drafting
> Run a department's decisions, delegation, and drafting from a Claude Project command center. No PHI, no clinical advice, human sign-off on everything.
**Author:** [Alex Lowe](https://theaicareerlab.com/about) — Founder, The AI Career Lab
**Published:** 2026-08-24
**Canonical URL:** https://theaicareerlab.com/blog/ai-chief-of-staff-for-physicians-2026
**Profession:** physician
**Category:** guide
**Tags:** physician, chief of staff, department chief, Claude Projects, executive assistant, delegation, 2026
---A department chief's calendar has two halves. One half is clinical and belongs to patients. The other half is a chief of staff's job that nobody hired a chief of staff for: forty open decisions, a delegation list that lives in your head, and a queue of memos, committee briefs, and policy drafts that all read "due last week."

The short answer to "can Claude be that chief of staff?" is: yes for the second half, never for the first. Set up a Claude Project as an administrative command center, run three weekly loops through it, and keep every patient and every clinical judgment out of it entirely. The setup below uses the command-center pattern from the [Executive Assistant AI Office](https://clowealex.gumroad.com/l/executive-assistant-ai-office?ref=physician-chief-of-staff) pack.

## What an AI chief of staff actually does

A human chief of staff does not make your decisions. They make sure every decision has an owner, a deadline, and the one-page brief it needs, and they chase what you delegated so you do not have to. That is exactly the job Claude can hold for the administrative side of a department.

Concretely, the AI chief of staff owns three registers:

- **Decisions** — every open call (the locum contract, the call-schedule change, the equipment request) with the options, the deadline, and who needs to weigh in.
- **Delegation** — what you handed to whom, when it is due, and when you last heard about it. The EA pack calls this the "waiting-on" list.
- **Drafting** — the memos, committee briefs, staffing justifications, and policy revisions that are administrative writing, not clinical writing.

What it does not own: anything about a patient, and anything that is a clinical judgment. The line is drawn in the next section and it is not negotiable.

## What it must never touch

This setup exists for the non-clinical half of the job. These rules are the price of using it at all:

- **Protected health information never goes into a prompt.** Not names, not MRNs, not dates of service, not "the 54-year-old in bed 12." Claude.ai and Cowork are not your EHR and are not covered by your organization's business-associate agreements unless your compliance office says so in writing. If a task needs PHI, do it in an approved system.
- **No diagnosis, no treatment, no interpretation of results.** The command center is an administrative drafting and tracking aid. It is not a clinical decision-support tool and must not be used as one.
- **You decide and you sign.** Review every output before it goes anywhere, and check policy language, regulatory references, and numbers against the source — model text is confidently wrong more often than it looks.

## Set up the command center

The command center is a Claude Project whose custom instructions hold your context, plus three living files the Project keeps for you: a fixed-layout home screen, identical every time you open it, with today's meetings, open follow-ups, the waiting-on list, and a numbered menu into the loops.

**Step 1: Create a Project** called something like "Department Office." (Cowork users set the same context once at the start of a Cowork task; the workflow is identical.)

**Step 2: Write the profile.** Custom instructions along these lines:

```xml
You are the administrative chief of staff for the chief of [department].
<context>
- Department: [size, sites, reporting line]
- Standing meetings: [M&M committee, division leads, hospital ops, etc.]
- Direct reports and what each owns: [list]
- Communication style: [brief, direct, formal with administration]
- Escalation rules: [what always comes to me; what my administrator handles]
</context>
<rules>
- Never accept or store any patient-identifying information. If I paste any, stop and tell me to remove it.
- You draft; I decide and sign. Mark every output DRAFT.
- Track every delegated item with owner and date.
- When unsure, ask one question rather than guess.
</rules>
```

**Step 3: Seed the three registers.** Paste your current open decisions, your delegation list, and your draft queue once. From here on the loops maintain them.

**Step 4: Ask for the home screen.** "Render the command center" should return the same layout every time: today's schedule with a prep note per meeting, the open-decisions table, the waiting-on table, and the menu.

## Three weekly loops: decisions, delegation, drafting

For a chief whose clinical days are unpredictable, three weekly loops fit better than daily ones.

**The decisions loop (Monday, 15 minutes).** Open the command center and ask: "Show every open decision with its deadline, the options on the table, and what input is still missing." Make the calls you can make. For the rest, have Claude draft the one-line request to whoever owes you input. The register updates; nothing is closed until you say so.

**The delegation loop (Wednesday, 10 minutes).** "Walk the waiting-on list. For anything past its check-in date, draft a short, warm nudge in my voice." You review each nudge, send the ones you agree with, and mark what came back. This is the loop most chiefs skip when they run it in their head, and it is the one that keeps a department from stalling.

**The drafting loop (Friday, 30 minutes).** Hand Claude the week's writing: the staffing justification for the CMO, the committee brief, the revised on-call policy. Give it the facts, the audience, and the decision you already made; ask for a draft, then a tighter draft. You edit and you sign.

## Where to start

If you want to build it by hand, the free [Claude Cowork for Executive Assistants](/resources/claude-cowork-executive-assistant) guide walks through the Project setup and the four core workflows; adapt the profile to a department rather than an executive. If you would rather start from a working office and edit, the [Executive Assistant AI Office](https://clowealex.gumroad.com/l/executive-assistant-ai-office?ref=physician-chief-of-staff) ships the command center, loops, and registers pre-built. And for the clinical-adjacent use cases that *are* appropriate, our [AI for physicians](/professions/physician) page covers documentation support, literature summaries, and patient-education drafting with the same no-PHI rule.

*This article is general guidance for physicians in administrative leadership. It is not medical, legal, or compliance advice; HIPAA, your organization's policies, and your compliance office govern actual AI use in your department.*
## Frequently asked questions

### Can I put patient information into the command center?

No. Nothing in this setup touches protected health information. The command center holds your calendar, your open decisions, your delegation register, and administrative drafts. Patient names, MRNs, dates of service, and anything else that identifies a patient stay inside your EHR and your organization's approved tools. If a task needs PHI to be useful, it is not a task for this setup.

### Is this a clinical decision-support tool?

No. It is an administrative drafting and tracking aid for the non-clinical half of a department chief's job: schedules, staffing memos, committee prep, policy drafts, and follow-up. It does not diagnose, recommend treatment, or interpret results, and it should never be used that way. Clinical decisions belong to licensed clinicians using validated tools.

### Do I need the paid Executive Assistant pack, or can I build this myself?

You can build it yourself with a Claude Project and the free Claude Cowork for Executive Assistants guide. The paid pack gives you the fixed-layout command center, the three loops, and the waiting-on and action registers pre-built, so you edit the profile instead of writing the system. Either way, you and your admin team still own every decision and every sent document.

---

*Canonical version: https://theaicareerlab.com/blog/ai-chief-of-staff-for-physicians-2026*
*This document is the Markdown companion served for AI crawlers and answer engines. See the canonical URL for the rendered version with navigation, related content, and interactive elements.*