← Executive & Administrative VA Lesson 6 of 7 · ~5 min

The AI layer: judgment is the job

AI is coming for the job you were about to apply for

Let's not soften it: scheduling, inbox triage, meeting notes, and routine drafting — the four pillars of entry-level admin work — are exactly what AI now does cheaply. The US BLS projects the role to decline because of it. But the same data shows the AI-fluent EA earning a 15–20% premium over the generic one. Same job title, opposite trajectories. The difference is whether you use the tools or compete against them.

Key idea: Stop doing triage and start designing the systems that triage. AI executes; you own the judgment, the gatekeeping, and the confidentiality.

What you hand to AI

ChatGPT / Claudefirst drafts of emails, summaries of long threads, research briefs

Fireflies / Ottermeeting transcription + AI summaries you distill into action items

Zapier (+ AI actions)the glue: form → CRM → task board → Slack, no human touch

Gemini in Gmail / Copilotinline drafting and thread summaries inside the client's stack

The workflow shift is concrete. Old way: read a 40-message thread for 20 minutes, then draft a reply for 15. New way: AI summarizes the thread in 30 seconds, you verify against the two messages that matter, AI drafts from your three-bullet instruction, you edit for accuracy and voice, done in 6 minutes. That's not cheating. That's the throughput your competition already has.

What you never hand to AI

Three things stay yours, and they're precisely what the premium pays for:

Judgment — AI can draft the reply to the angry client; it cannot know that this client threatened to leave twice this quarter and the founder wants to call him personally. Context lives with you.
Gatekeeping — deciding who gets the founder's time is a political act. AI will happily book a meeting with anyone who asks nicely.
Confidentiality — you are the human firewall around salaries, deals, legal matters, and personal messes. AI has no concept of "this must never leak."

Reality check: Pasting client data into your personal ChatGPT account can violate your NDA and get you terminated — some free-tier tools train on what you paste. Use client-approved accounts only, redact names and financials when possible, and ask about AI policy in week one. Asking marks you as senior; assuming marks you as a risk.

Build a prompt library

Amateurs type a fresh prompt every time and get inconsistent output. Seniors keep a library of tested, reusable prompts for recurring work — EOD polish, thread summaries, first-draft replies, research briefs. Here's the shape of a good one:

Prompt template — inbox thread summary

"You are an executive assistant. Summarize this email thread for a busy founder in exactly this format: SITUATION (2 sentences max), WHAT THEY WANT (1 sentence), DEADLINE (date or 'none stated'), RECOMMENDED ACTION (1 sentence, start with a verb). Do not soften bad news. Flag anything involving money or legal terms. Thread: [paste]"

Notice the moves: a role, a rigid output format, explicit rules ("do not soften bad news"), and a flag condition. Every prompt in your library gets refined each time the output disappoints you. Within a month you own a toolkit no job post can teach.

The supervision habit

AI output is a draft from a talented intern who occasionally lies with total confidence. Before anything AI-touched leaves your hands, run the check: names and numbers verified against the source, no invented facts or dates, tone matches the relationship (AI defaults to corporate-cheerful, which reads fake from a trusted EA), and nothing confidential embedded. Mika in Quezon City caught Fireflies attributing a $50k budget approval to the wrong person in a meeting summary — that 30-second check is the job now.

Pitching the AI-augmented EA

This is where the 15–20% premium gets claimed — in how you sell it. Not "I know ChatGPT" (everyone says this, it means nothing). Instead:

Say this — in your application or interview

"I run an AI-augmented workflow: transcription and drafting are automated, which means I return summaries within an hour of your meetings and your inbox runs at zero in half the usual hours. The time saved goes into building your SOPs and automations. You're not paying me to type — you're paying me to run the system and make the calls AI can't."

Price the system you install, not the hours you sit. A founder paying $9/hr (~₱520/hr) for that pitch is getting more output than $6/hr of manual typing — and they know it.

Do this now

Build the first five prompts of your library: (1) email thread summary, (2) first-draft reply from bullet points, (3) meeting-notes-to-action-items converter, (4) research brief on a company, (5) EOD report polisher. Test each on real material — an actual thread from your inbox, real notes. Refine until the output needs under one minute of your editing. Save them in a doc titled "Prompt Library v1" with a one-line usage note per prompt. This doc is portfolio material and interview ammunition. 25 minutes.

Tip: use your ← → arrow keys.