How chiefs of staff actually use AI in 2026

Chiefs of staff are the most interesting AI users in operator-land right now. They're closer to the work than executives, more cross-functional than ICs, and they live or die by how much they can compress on a given day. Every one I've ever met is on a constant hunt for leverage.

We talked to 18 of them. The patterns were sharper than I expected.

What's actually in the daily stack

Across the 18 interviews, three tools came up nearly universally:

1. 1An AI inbox layer (Triagd, Shortwave, or — surprisingly often — a custom Zapier + Claude API setup)
2. 2A meeting note-taker (Granola, Fathom, or Otter, with Granola winning on stickiness)
3. 3Claude (or a writing layer over it, like Cursor or Lex) for drafting docs, decks, and one-pagers

Everything else was idiosyncratic. Some used Reflect for personal notes, some used Notion AI, some used nothing. But the inbox layer, the note-taker, and the writing tool were the bedrock.

What changed in the last 12 months

The biggest delta from 2025 to 2026 is that chiefs of staff stopped using AI as a one-shot tool and started using it as a continuous layer. In 2025, they'd paste something into ChatGPT, get an output, and copy it back. In 2026, the AI is just there — reading their email, sitting in their meetings, suggesting next sentences as they type. The friction tax of switching apps is gone.

Quote from one of our interviews:

“I don't "use AI" anymore. I just work. The AI is already there, drafting, summarizing, prepping. The asks I make of it are smaller and more frequent. I think I touched ChatGPT directly twice last month.”

Which workflows compound

Three workflows showed up repeatedly as ones that had become 10x leverage moments rather than nice-to-haves:

1. The morning briefing

Universally: a structured morning summary of overnight email, calendar conflicts, and outstanding decisions waiting on the CEO or themselves. Some had this from Triagd, some from custom scripts. Format varies, intent doesn't — they want the day pre-loaded before they open Slack.

2. The recap loop

After every meeting, the note-taker generates a recap. The chief of staff edits it (10–60 seconds), then forwards to attendees with action items broken out. This workflow had the highest reported satisfaction of any AI use case in our interviews. It's small, it's repeatable, it compounds: a year of perfect recaps is a year of unambiguous accountability.

3. The decision document

Before any non-trivial decision, a chief of staff drafts a 1-page memo. In 2026, that memo is co-authored with Claude: "Here's the context, here are the options I've heard, here's what I think we should do. Tighten it." The AI is the editor, not the author — but the editor cuts 30 minutes per memo, and chiefs of staff write a lot of memos.

What's hype

We asked specifically about AI tools the interviewees had tried and dropped. Three categories came up repeatedly:

• AI scheduling assistants. "They're worse than just texting the other person," was the consensus. The back-and-forth confidence is too low. Chiefs of staff still book meetings manually.
• AI CRM enrichment. Hit-and-miss accuracy, and the misses are public-facing (you reference a fact about a partner that's wrong, in writing, and it's embarrassing). Most have churned off the standalone tools and use whatever's bundled into HubSpot or Attio.
• "AI agents" that take multi-step actions autonomously. Almost universal skepticism: the chiefs of staff want AI that drafts and summarizes, not AI that acts. The risk:reward isn't there yet.

The honest pattern

Across the 18 conversations, the most successful AI adopters had one thing in common: they treated AI as a writing and reading layer, not a doing layer. The tools they kept were the ones that helped them read faster (briefings, summaries, recaps) and write faster (drafts, memos, replies). The tools they dropped were the ones that tried to act on their behalf.

That's worth holding onto. The hype right now is all about AI agents that take action. The actual best practice from the operators who are getting the most out of AI right now is the opposite: keep the human in the loop, let the AI prepare the work, you press send.