Gemini Cloud Agent Interview: One Memory Across Your Full Stack
Read the Portuguese translation →
Author: Google Cloud · Watch on YouTube
Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.
Original language: English
Original video: Gemini Cloud Agent Interview: One Memory Across Your Full Stack · Channel: Google Cloud · 02:10 · Watch on YouTube
Topics: AI agents, Task and workflow automation, Prompt and context engineering, AI infrastructure and hardware
Summary
Section titled “Summary”Adam Fisher describes how Gemini agents make agent capabilities available across the enterprise, including to non-developers. Gemini can learn a person’s memories and preferences, then use scheduled routines to work in the background. Cloud-based agents can run across tools such as GWS, Microsoft, and Slack, even when the user is offline. Gemini can also create agents for multi-step workflows from a person’s instructions and tailor its outputs using preferences recalled through dream mode.
Key points
Section titled “Key points”- Gemini agents bring agent capabilities to people across enterprise roles, not just engineers.
- Gemini can build memories and preferences based on how a person works.
- Users can set routines and schedules that run in the background.
- Cloud deployment lets agents run wherever employees work, including GWS, Microsoft, and Slack.
- Agents can continue working on a user’s behalf when their laptop is closed.
- Gemini can automatically construct agents from a person’s words for complex workflows.
- Skills are used to get the requested work done.
- Dream mode extracts memories and preferences from conversations for future use.
- Gemini can tailor artifacts based on preferences such as communication and writing style.
Chapters
Section titled “Chapters”| Start | Chapter | Summary |
|---|---|---|
| 00:00 | Agents for everyone | Fisher explains that Gemini agents make enterprise intelligence accessible to people in any role, not just engineers. |
| 00:34 | Routines and cloud orchestration | Gemini can use memories and preferences to run scheduled routines in the background and work across cloud-based tools. |
| 01:05 | Agents for complex workflows | Fisher describes how Gemini can build agents from a person’s instructions and use skills to complete multi-step work. |
| 01:38 | Learning through dream mode | Dream mode draws memories and preferences from conversations so Gemini can tailor its outputs to the user. |
Quotes
Section titled “Quotes”With Gemini agent, that power comes to everybody in the enterprise. — 00:00
I can create routines and schedules that run for me in the background — 00:34
the agent can run even when I’m not working. — 01:05
Gemini builds agents automatically using just a person’s words — 01:05
Gemini will pull out through dream mode the memories and preferences from our interactions. — 01:38
Concepts and entities
Section titled “Concepts and entities”Gemini, Gemini Cloud Agent, dream mode, sub-agents, skills, Adam Fisher, Google Cloud
Full transcript (original language)
Section titled “Full transcript (original language)”00:00 I’m here with Adam Fisher who just gave a great demo on how Gemini orchestrates complex workflows. Adam, you talked about the power of agents and how that unlocks things for non-developers. Could you tell us more about that? >> Yeah. For the last year, engineers have been having all the fun. With Gemini agent, that power comes to everybody in the enterprise. So, regardless what your role, you can now access this intelligence and take it with you where you work. >> Amazing. And how are you using this power? Is there anything that it’s unlocked for you? >> Yeah, for me, this starts a lot with a scheduled routines. So, now that Gemini knows who I am, how I work, it’s able to
00:34 build the right memories and preferences. I can create routines and schedules that run for me in the background while I’m focusing on other things. >> That’s so helpful. Can you also tell me a bit about how Gemini in the cloud helps with orchestration? >> For a long time, engineers have been using intelligences on the desktop. Bringing it into the cloud means that we can run it anywhere employees work. So, if I want to have Gemini available in GWS or in Microsoft or in Slack, I have that available and the agent can run even when I’m not working. So, even when
01:05 my laptop is closed, it can keep working for me on my behalf in the background. >> And tell me more about how sub-agents play into this and how they help with the complex multi-step workflows. >> We’re moving from a world where you had to build your own agents. And there’s a lot of tools that we’ve provided for a long time to set up runtimes and give different model preferences, even things like workflow builders, right? Where you can set step-by-step guidance for agents. That world is changing. Instead of needing to be really explicit in how these agents work, Gemini builds agents automatically using just a person’s words to construct exactly the agents that are needed and then using skills to
01:38 get that work done. >> And I also remember you talking about dream mode. Can you explain how Gemini learns through dream mode? >> In our conversations, Gemini will pull out through dream mode the memories and preferences from our interactions. And then it can reuse those to thumbprint to fingerprint to my preferences. So, it can use things like how I communicate or my writing styles to give me back the artifacts that are most helpful to me. >> Amazing. So, it’ll keep learning about me so that as it like helps me do tasks, it’s going to be more tailored to what I need. >> Exactly. Exactly. >> Amazing. All right, thank you so much,
02:10 Adam. This has been great. >> [music]