Skip to content

Session 11

Thursday, 22 October 2026 · 1 hour

Add to Google Calendar

The recurring AI Office Hours event covers every session — adding it once gets you all of them.

Eleventh AI Office Hours. Two demos: Josh Sheppard walks through AI training available through Coursera and how to get started, followed by Rob Bowron on building a Chief of Staff automation in Workspace Studio.

Run of show

Time Beat
0:00 Welcome + recap — what happened since Session 10, what's new on the hub
0:03 Review the Solution Library + Ideas queue — new entries, status changes, what's been triaged
0:13 Demo 1 — Josh Sheppard — AI training through Coursera — what's available, how to access it, and where to start
0:33 Demo 2 — Rob Bowron — Building a Chief of Staff in Workspace Studio — how to set up an automated assistant that surfaces what matters from your email, calendar, and chats
0:53 Open mic / Q&A
0:58 Topics for next session — what to cover in Session 12
1:00 Hard stop

Pre-read (optional)

After the session

Rob's Chief of Staff build is now a self-serve kit: CoS Briefing Build Kit — code, skills, and setup steps, taught module by module in Sessions 12–16.

Session recap

Opening — AI news & context (0:00–0:12)

Rob opened with personal agent updates: a friend's agent successfully coordinated with Delta Airlines customer service to locate and return misplaced AirPods. Mariano Franco noted that AI recently solved several Clay Institute mathematical problems and advanced physics challenges — causing real concern among researchers about job security. Adi Kumbagiri shared that using two personal agents for three months has saved significant time negotiating outside commitments. Brett Salyer raised a counterpoint: AI agents are increasing customer care call times and creating potential denial-of-service issues, prompting Aditya to flag personal agent protocol standards as a future agenda item.

Q&A

Q: What is Coursera offering and how do we access it? (Josh Sheppard) A: Scotts Miracle-Gro has contracted 1,000 Coursera licenses for a Generative AI skills track covering 1,100+ AI courses including Google and Gemini content. Licenses only count against the limit when actually used. Learning paths integrate directly with Workday — assignments will appear as task notifications on your homepage. Go-live target is December 1. A pilot team is forming now for newer AI users; contact Josh Sheppard to join. The courses feature embedded LLM functionality directly in the interface.

Q: What does the Chief of Staff build actually do — can you show the architecture? A: Rob walked through the full build: a Google Apps Script project ("chief of staff gather 2") deterministically pulls calendar, drive, and email data via your Google ID. That data feeds into a multi-agent Workspace Studio flow — an analyst skill categorizes meetings using a RACI-style framework, checker agents verify outputs against source data and remove unsupported claims, and a writer agent produces the final briefing. Web and Workspace search are disabled on the analyst step intentionally so the AI only processes what the script fetched, not hallucinated external data. The build kit is available on the hub: CoS Briefing Build Kit.

Q: Gems vs. Skills — what's actually different? A: Skills replace Gems in Workspace Studio. Key practical difference: Skills handle longer instruction sets much better. Current limitation: Skills don't parse uploaded knowledge files as effectively as Gems did. Michael Lockwood has unlocked document-sharing permissions across SMG so everyone can access the build kit. A dedicated session on converting Gems → Skills is planned.

Q: Can you use both Claude and Gemini together in a workflow? A: Melissa Bond noted their team already does this — using Claude and Gemini back-to-back to improve outputs for process flow maps and program troubleshooting. AI has removed the technical barriers of Google Apps Script and Sheets for non-developers.

Q: Why does Gemini give different answers across surfaces? A: LLMs are probabilistic, not deterministic — different agent instantiations across surfaces produce varied nuances. Rob recommended using checker agents to validate important outputs before acting on them.

Next steps called out in the session

  • Rob to share the Chief of Staff build kit link on the CoE site (done — see CoS Briefing Build Kit)
  • Associates interested in the Coursera pilot: contact Josh Sheppard
  • Aditya + Brett to follow up on personal agent protocol standards
  • Follow-up session planned to help participants build their own multi-agent workflows
  • 15-minute office hours available — book time with any CoE team member from the hub homepage