Back to blog
FILE 0x66·THE AI CHIEF OF STAFF IS A JOB TITLE WAITING TO BE INVENTED

The AI chief of staff is a job title waiting to be invented

June 7, 2026 · ai, product, founders, saas

I've been running my own AI chief of staff for about a year. The short version: it works. The longer version is why it doesn't exist as a product yet.

What a chief of staff actually does

Human chiefs of staff — the good ones — do three things. They surface what the principal needs to know before the principal asks. They handle the low-signal work so the principal can stay in the high-signal work. And they maintain the running context so every handoff is warm.

That third one is the reason most AI assistant products fail in this role. They're stateless. Every conversation starts fresh. You're not paying for a chief of staff — you're paying for a very fast transcriptionist.

The access problem

Every AI product I've seen either lives in an app or lives in a browser extension. Neither of these are where executives actually work. They work in their texts and their inbox.

I don't want to open an app to ask whether I have a conflict Thursday afternoon. I want to text the same way I'd text a person who already knows my calendar.

The infrastructure for this exists — Twilio has phone numbers, Signal has an API, Claude has 200k-token context. Nobody has assembled it with the right UX wrapper.

The context problem is actually a setup problem

Most AI tools are sold on the premise that the AI will figure out who you are through conversation. This is technically true but practically too slow for an executive use case. If I'm on the road and I need a quick answer about a client, I need the AI to already know who that client is, what our relationship is, and what the live issue is.

That knowledge doesn't emerge from the AI. It gets seeded by someone who takes the time to brief it properly. The setup is the product. A 90-minute onboarding session where you walk through your team, your key accounts, your calendar rhythm, and your preferences — that turns a general-purpose model into something that actually knows you.

The cost of that setup is why this doesn't scale to $20/mo. But it does scale to $500/mo for the right audience.

What I built

My assistant, Cass, can:

None of this is technically impressive in 2026. The hard part was building the memory layer, deciding what to persist and how to keep it current, and wiring the right access to the right channels.

The result is an assistant I actually use. Not for everything — but for the moments where I need a fast briefing or a first draft, it's faster than doing it myself.

The productization gap

The challenge with productizing this is the setup cost. Every new client needs:

That's not zero labor. It's maybe 2-3 hours upfront and an hour a month. At a $500/mo price point, the labor is a meaningful cost. At $1,500/mo, the margin is fine.

But the distribution is the real constraint. Who buys this? Founders. Executives. People whose hourly opportunity cost is high enough that $1,500/mo is a rounding error if it saves them an hour a week.

That's a small, specific market. Which is fine — you don't need a million customers to build a meaningful business at $1,500/mo.

Why this isn't just a ChatGPT wrapper

The honest answer is: it mostly is, in the sense that Claude is doing the language work. The value-add is:

  1. The channel. SMS/Signal, not an app.
  2. The persistence. Memory that survives across sessions.
  3. The integrations. Real calendar, real email.
  4. The setup. Someone who briefs the AI properly at the start.

Item 4 is the most underrated. The reason most AI assistants feel generic is that they haven't been told anything. The product isn't the model — it's the context and the curation.

I'm taking ten beta clients for a productized version of this, called Evangeline. If you're a founder or exec who's been underwhelmed by AI assistant products and want to try something that's been running in production for a year, reach out.

The waitlist is at withevangeline.com.