Chief of Staff and executive assistant roles are quietly being re-scoped around AI agents. Granola, Fireflies, and Fathom for meeting intel. Superhuman AI, Fyxer, and Shortwave for inbox. Motion and Reclaim for calendar. ChatGPT and Claude Projects for briefing memos. Increasingly, agentic browser tools for booking and research. The role now commands $150-250K at YC growth-stage companies and Fortune 500 executive offices. And there is no consolidated, opinionated playbook for the full CoS AI operating system.
This hub is for the Chief of Staff, executive assistant to C-suite, business ops lead, founder's associate, or CEO office operator. It curates the twelve-agent operating system — meeting recorder, inbox triage, calendar defender, briefing memo drafter, research runner, action-item tracker, contact CRM, expense reconciler, travel planner, doc drafter, presentation builder, and status reporter — that together let a single competent CoS operate at three-person capacity. Every skill is battle-tested, tool-specific, and honest about what still requires human judgment.
Who this hub is for
Chief of Staff, executive assistant to C-suite, business ops lead, founder's associate, CEO office at Fortune 500 and growth-stage startups.
Meeting intel (Granola or Fathom), inbox triage (Superhuman AI or Fyxer), calendar defender (Motion or Reclaim), a general assistant (ChatGPT Enterprise or Claude Projects), one search-across-your-stuff tool (Glean or Notion AI), one contact-relationship layer (Clay or Attio). Six tools. Learn them deeply. Add the seventh only when you have a specific gap.
Granola wins for user experience — no bot in the meeting, transcript-plus-notes UX, offline-first. Fireflies wins for enterprise features and CRM integrations. Fathom wins for free-tier value and coaching-focused analytics. Chief of Staff use most often defaults to Granola for its "no bot, no awkwardness" pattern with executives who dislike a visible recorder.
Yes for triage and drafts, no for send-without-review. Superhuman AI and Shortwave both draft in your exec's voice from example emails. You review, edit, send. Sending an AI-drafted email without review is a career-ending risk if the AI hallucinates a commitment or misreads context. Save 70% of the drafting time, keep 100% of the send discretion.
Standard four-part structure: attendee context, meeting objective, discussion points, likely asks. Feed the exec's calendar entry, the counterparty's LinkedIn and company profile, the last three emails with them, and any relevant Slack threads into Claude Projects. Ask for the four-part memo. Review, add human judgment on likely emotional or political dynamics, deliver 24 hours before the meeting.
Agentic booking uses a browser-driving AI to book restaurants, hotels, and appointments on your behalf. As of 2026 the technology is functional but not reliable — 70-85% success rate depending on site complexity. Use it for low-stakes bookings (dinner reservation, flight seat selection). Do not use it for high-stakes bookings (a Michelin reservation for an anniversary, an executive-team offsite). Human review remains critical.
The role shifts from execution ("book the meeting, take the notes, follow up") to orchestration ("design the agent that books, records, follows up; then handle the exceptions"). Successful CoS operators in 2026 spend more time on judgment calls, relationship management, and strategic prep — the work that AI cannot yet do — and less time on the mechanical work AI absorbed. Compensation reflects the shift.