Session 5
Thursday, 30 July 2026 · 1 hour
The recurring AI Office Hours event covers every session — adding it once gets you all of them.
Fifth AI Office Hours. Two associate demos on the AI-as-personal-programmer pattern — one on turning a Google Sheets DB into a mobile-friendly HTML tool, one on AI-generated executables for data-quality work. Walk the latest Solution Library and Ideas queue updates, walk the next AI Brief digest.
Run of show
| Time | Beat |
|---|---|
| 0:00 | Welcome + recap — what happened since Session 4, what's new on the hub |
| 0:03 | Review the Solution Library + Ideas queue — new entries, status changes, what's been triaged |
| 0:13 | Associate demo — confirmed — Travis Askew (Sales) on turning a Google Sheets DB into a mobile-friendly HTML tool — uses his sales-contest app as the worked example: standard Google Sheet becomes a phone-friendly interface for the field. Same pattern reusable for any team that lives in Sheets but needs a frontline UI |
| 0:28 | Associate demo — confirmed — Christian Myers (Master Data) on AI-generated executables — used Gemini Gems to design the architecture and write the Python that compiles to a .exe for checking Product Hierarchy quality issues. A layer deeper than Mathieu's pattern: Mathieu wrote AppScript inside Workspace; Christian shipped a standalone compiled tool |
| 0:43 | Latest in enterprise AI — walk the top AI Brief pick from the new fortnight digest |
| 0:48 | Open mic / Q&A |
| 0:58 | Topics for next session — what to cover in Session 6 |
| 1:00 | Hard stop |
Pre-read (optional)
- How to Participate — ground rules and norms
- Solution Library — what's been added since 2026-07-16
- Ideas — current queue and triage notes
After the session
Q&A distilled from what came up live — the takeaways most relevant to associates.
What did Travis Askew demo?
Travis (Sales Operations) showed how he turns standard Google Sheets into mobile-friendly HTML apps using Google App Scripts and Gemini. The Sheet is the database; App Scripts generates a URL-accessible frontend. Key features: auto-provisioning by email so each user only sees their own data, a full access hierarchy (sales manager sees their team, district manager sees their district), an impersonation mode for troubleshooting others' views, and weekly automated email updates triggered on a schedule. He builds from a personal set of "build files" — reusable documents covering brand standards, security parameters, and mobile UI conventions — so he's not starting from scratch each project. Travis is open to sharing these; reach out directly if you want a copy.
What are the limits of using Google Sheets as a database?
It works well for small to medium datasets. Travis hit the wall at roughly 10 million cells on a CRM tool. Workarounds: split data across multiple Sheets and reference each by URL, use a scheduled trigger every few minutes to cache data and keep the app responsive, or query only the data relevant to a given view. For very large or heavily queried datasets, consider moving to a different platform. Keeping unused cells cleared also helps — empty cells still get queried.
What did Christian Myers demo?
Christian (Master Data / Data Governance) used Gemini Gems to write a Python executable that automates product hierarchy validation — comparing the master file in Google Drive against SAP backend tables. What used to take roughly 40 hours a month now runs in under 5 minutes. The output is a structured report with four tabs: values missing from SAP, orphaned SAP values that shouldn't exist, spelling discrepancies between the two systems, and materials with incomplete or outdated hierarchy assignments. The app includes error logs, audit logs, and a readme so any teammate can pick it up without needing Christian there to explain it.
How do you manage Gemini's forgetfulness in long conversations?
Two approaches came up. Travis creates "build files" — documents capturing his security parameters, personas, brand standards, and UI conventions — and loads them at the start of every project so the model has context from the first message. Christian uploads his Python scripts and datasets directly into a Gem, where they persist across conversations. Both approaches avoid the failure mode of a long chat thread where the model loses track of earlier decisions. The practical rule: anything you'll repeat across sessions belongs in a file, not a prompt.
Any tips for getting better results out of Gemini?
Talk instead of type — Travis found voice input captures nuances that get dropped when typing. Start with your build files loaded before opening a conversation. Be explicit about constraints: security, brand, mobile-first, data source limits. If you land on a result you like, document the parameters that produced it immediately. Christian adds: put reusable scripts in the Gem's knowledge base so the model refines them over time rather than recreating them from scratch each session.
Security: what should I know before running AI-generated scripts?
Brett (Security) flagged it directly: you are responsible for the code that runs on your computer. Ask Gemini to explain every script before you run it — and use that explanation to audit whether the script is actually doing what you want. This matters most as scripts move from personal productivity into anything that touches corporate systems, shared data, or external connections. Review authentication carefully before promoting a personal project to something broader. If you trigger a security alert, the security team will ask what you're running.
Are tools other than Gemini approved for use at Scotts?
Gemini is universally available — no approval needed. Claude and Cursor (which supports Claude and other LLMs) are available via a restricted use case request through ServiceNow. Any tool that involves uploading data externally requires a legal agreement to be in place first. Office hours are deliberately Gemini-focused because it's the tool everyone has access to; a separate smaller session on Claude best practices may be worth exploring for users who already have approval.
How is the CoE tracking project requests going forward?
Monday.com is being decommissioned. Requests will move to ServiceNow — consistent with how other IT tickets are handled — feeding into existing Jira boards. The CoE already has a Jira board in place. More detail on the new process will be shared once it's set up.
When's the next session?
Thursday, 2026-08-13. Same recurring calendar event — add it once from any session page and you've got the full series.