What This Google Employee Learned About AI Agents That You Should Know
Stay updated with insights on AI agents from a Google employee.
Don't miss any news!
She has access to AI tools that are not yet widely available: Lisa Ihde has been working as a Technical Account Manager at Google for over two years. In the t3n podcast "Work in Progress," she shares how she effectively uses AI agents, where they hallucinate, and how she verifies results.
She acts as a liaison between Google customers and internal engineering teams: Lisa Ihde works as a Technical Account Manager, or TAM, at Google. As a writer and speaker, Ihde has early access to new AI models and tests them in her daily work.
AI Agent Takes Over Feature Requests
She discusses how her role has recently changed due to AI agents in the t3n podcast "Work in Progress." As an example, she brings up a recurring task. Ihde needs to collect feature requests, which are customer requirements for software. Previously, this was a manual task, and the companies' requests had to be manually entered into an Excel spreadsheet.
Today, Ihde has built workflows that automatically go through tickets and send her weekly reports. She mentions hundreds of requests per customer that are now sorted automatically.
AI Struggles with Task Summaries and Abbreviations
Instead of relying on the results, she verifies them. Ihde has noticed how easily unpleasant errors can occur when working with AI agents. For instance, automatic meeting notes sometimes lack updates. The AI may recognize the initially mentioned task but struggles to capture when a new status is discussed in the meeting that requires a task adjustment.
Additionally, Ihde noticed that Gemini sometimes has trouble with abbreviations. Internal terms are mis-mapped, and sometimes the AI even hallucinates product names. "A colleague pointed this out to me. I hadn’t even noticed it," she shares. Since then, she pays closer attention.
Publish Content Only After Review
That’s why no content goes out automatically for her. A mistake with feature requests could be fatal: "Imagine if the agent finds an edited field somewhere and tells the customer that the feature will be released in a month – when that was just a mistake."
But does Ihde feel her role is threatened? After all, AI agents are already taking over some of her tasks. How she deals with this and how she views the future is shared in the linked podcast episode.
Listen to the t3n Podcast "Work in Progress"
You can easily subscribe to the t3n podcast "Work in Progress" in your preferred podcast app. Typically, you can find the podcast by simply entering "t3n Work in Progress" in the search. Alternatively, you can also manually enter the RSS feed in the app.
- RSS Feed for t3n Work in Progress
- t3n Work in Progress on Apple Podcasts
- t3n Work in Progress on Spotify
This text was created using our internal editorial AI tool.
This article was originally published on June 21, 2026, but it still interests many of our readers. Therefore, we have updated it and made it available again.
