Hello everyone,
Over the past several weeks, we’ve explored how Executive Assistants can use AI to write better, build reusable AI assets, connect tools, prepare executive context briefs, and catch mistakes before they slip through the cracks. This week, we’re taking the next step.
Instead of asking AI to complete one task at a time, what if you could give it an outcome, provide the right context and guardrails, and let it work through several steps before coming back to you?
That’s the idea behind agentic AI.
And while fully autonomous AI agents are still not appropriate for many EA responsibilities, the shift is already underway. Most organizations are still experimenting rather than deploying agents at scale, but workplace AI is steadily moving from one-off chatbot interactions toward multi-step workflows that operate with human oversight. For EAs, that makes this less about learning another AI tool and more about learning a new skill: How do I delegate to AI without giving up control?
At a glance
💡 Key takeaway: Start identifying repeatable workflows where AI can gather, organize, analyze, and prepare work for you, while you remain responsible for the decisions that matter.
🎯 Level: Advanced
⏱️ Time to read: 8 - 10 minutes
💻 By the end: You’ll be able to distinguish a prompt from a workflow and an agent, identify where human approval belongs, and map one recurring EA responsibility into a semi-agentic workflow.
AI Term of the Week: Agentic AI
Agentic AI refers to AI systems that can work toward a goal across multiple steps rather than simply responding to one prompt. “Agentic” does not have to mean autonomous. For most EA work today, the sweet spot is semi-agentic: AI does more of the preparation while a human remains responsible for consequential actions.
Part 1: The Shift: From “Do This” to “Help Me Accomplish This”
Most of us started using AI like this: Prompt → Output
“Summarize this email thread.”
“Draft a follow-up email.”
“Turn these meeting notes into action items.”Those are useful tasks. But every time, you initiate the work, provide the information, receive the answer, and decide what happens next.
An AI workflow changes that. Instead of giving AI one isolated instruction, you design a repeatable process:
Trigger → Gather information → Process it → Prepare an output → Review
An agent takes the idea one step further. You give it an outcome, and the AI can determine some of the steps required, gather context, use connected tools, and know when it needs human input. Think about it this way:
Prompt: “Summarize these documents for tomorrow’s meeting.”
Workflow: “When documents are added to the meeting folder, summarize them using my briefing template and prepare a draft for me.”
Agent: “Prepare me for tomorrow’s meeting.”With sufficient access and instructions, an agent might determine what meeting you mean, review the calendar, locate relevant documents and communications, identify the attendees, summarize the context, draft your briefing, and then bring the result back to you for review.
Part 2: Turn Your Executive Brief Into a Workflow
Remember our Week 5 exercise on building a Friday Executive Context Brief? Instead of starting from scratch every Friday, imagine turning that into a recurring workflow.
Your goal could be: “Prepare a one-page briefing that gives my executive the context they need for the upcoming week.” Now let’s break the job down:
Gather: Review the upcoming calendar, important email threads, meeting notes, project updates, and other approved sources.
Identify: Determine the meetings, decisions, commitments, risks, and priorities that deserve the executive’s attention.
Synthesize: Connect information across those sources instead of simply summarizing each one separately.
Draft: Prepare the briefing using your established template, tone, and priorities.
Check: Flag missing information, conflicting details, uncertain conclusions, or anything requiring human judgment.
Pause: Send the draft to you, not your executive. You review it, correct the context, add what AI could not know, and decide what goes forward. Only then is the briefing ready.
Part 3: Design the Stop Signs
Here’s where advanced AI use becomes very different from simply writing advanced prompts. A good agentic workflow isn't only about deciding what AI can do. You also need to decide where AI must stop.
That’s called Human-in-the-Loop (HITL).
AI can probably handle:
Summarizing internal information, organizing documents, categorizing tasks, extracting action items, identifying patterns, and preparing internal drafts.
AI can prepare, but you should review:
Executive briefings, agendas, reports, proposed schedules, travel itineraries, stakeholder summaries, and draft communications.
AI should stop and ask before:
Sending an email, changing a meeting, modifying an important record, sharing confidential information, making a commitment, publishing externally, or taking another action in your executive’s name.
Part 4: Simple Rule for Deciding What to Delegate
When you created an AI asset library, documented your executive’s preferences, developed reusable prompts, or built an Executive Context Brief in previous weeks, you were building the context infrastructure that makes more advanced AI workflows possible. Before adding AI to any recurring responsibility, ask these questions:
1. If AI gets this wrong, how difficult is it to fix? If the mistake is internal and easily reversible, more automation may be appropriate.
2. Does this affect another person or make a commitment? If yes, AI should probably prepare rather than execute.
3. Does AI actually have the context required to make the decision? If it doesn't, either provide the missing context or keep that step human.
4. Is the task fundamentally about judgment, trust, relationships, or executive intent? If yes, you should own it.
Bonus: AI Updates of the Week
Microsoft 365 Copilot rolled out major August updates, including richer Cowork agents, new Meeting Recaps in Teams, and SharePoint dashboards with one‑click Copilot buttons for contextual prompts.
Google launched Gemini Spark, a persistent personal agent that keeps working on tasks after you log off and deepens Gemini in Drive for cross‑document Q&A and automation.
EU AI Act Article 50 and California SB 942 took effect, requiring stronger AI disclosure, provenance tags, and content labeling - pushing companies to formalize AI governance around workplace tools.
Final Thoughts
The future of AI-powered EA work is about becoming the EA who knows how to combine human judgment with AI capability. Your instincts, relationships, and understanding of your executive still matter. AI simply expands what you can accomplish with them.
See you next week!
Hit reply and tell us: What's your biggest challenge when managing your own inbox and your executive's? We'd love to hear from you and build future newsletters around the workflows that matter most to you. Your ideas may inspire a future edition of The Curve!
The Base Team
