Ben Thompson’s annual pre-Google Cloud Next interview with Google Cloud CEO Thomas Kurian covers agents, Gemini’s harness quality, TPUs, and how Google thinks about its integration advantage. The conversation was recorded on April 15, ahead of the keynote. Stratechery published it alongside the keynote broadcast.
Kurian joined Google Cloud in 2018, having spent 22 years at Oracle as President of Product Development. This is Thompson’s fourth interview with him, following conversations in 2021, 2024, and 2025.
What Kurian says changed from last year to this year
Thompson’s direct question — what has actually changed that makes agents real now, rather than theoretical — gets a structured answer. Kurian lists several changes.
The first is model capability: “Gemini is able to reason much more effectively as new versions of Gemini have come out.” The second is long-running memory: agents automating tasks across many steps need to maintain state, and the models can now do that more reliably. The third is tool interaction: “there have been good abstractions, skills, tools, MCPs [Model Context Protocol], as they’re called, they’re all abstractions for how an agent reasons and interacts with the rest of a company’s systems.”
Kurian summarized: “All of them have advanced and so the core capabilities that the models themselves have gotten a lot better, the capability and the ability to use tools and interact with the rest of the world has become a lot better, the abstractions that the world exposes itself to the model has improved.”
The keynote framing, which Thompson notes he watched after the interview, reinforced this: Kurian returned to last year’s theme of a unified architecture but “emphasized that the use cases were no longer theoretical or pilots but running at scale for real users.”
Gemini agents in production: letting customers speak
Thompson’s question on actual agent performance is pointed: he notes that Gemini was prominent in community discussion four months ago but that more recent attention has been on Anthropic and Claude, Codex, and that “Gemini, not much talk.” Kurian’s response is consistent with his standard approach: “Let our customers talk about it, rather than we talk about it.”
The specific customer examples he gives in the interview cover a range of enterprise use cases. Citi is using Google agents for wealth management — described as researching a client’s investment priorities against their portfolio to produce recommendations. Comcast is using agents for consumer services: scheduling appointments, dispatching field technicians, and managing repair flows with “very complex flows that have many, many steps.”
Whether these deployments involve Gemini models exclusively or draw from the 200+ models available via Model Garden is not specified in the interview excerpt.
Google’s integration advantage
A recurring thread in Thompson’s framing is whether Google’s integration advantage — running on the same infrastructure as Google Search, Gmail, Workspace — is paying off in enterprise contexts. Kurian emphasizes this in his keynote-framing answer as well. Thompson notes that both Kurian and Sundar Pichai emphasized at the keynote that “Google itself was running on the same infrastructure as Google Cloud,” and that “Google Cloud was running the same stack as Google itself.”
Kurian’s argument is that Google’s own deployment is the largest and most demanding customer of its agent infrastructure, and that this creates a tighter feedback loop for production hardening than a purely external customer base would provide.
AI and cybersecurity
Kurian flagged cybersecurity prominently in the interview — Thompson notes the point came up in the keynote as well. The concern is that AI accelerates the speed of cyberattacks, and Google’s response is to bring AI and security capabilities together, including through the integration of Wiz. This is positioned as a feature of the enterprise agent platform rather than a separate product line.
Thompson’s observation after watching the keynote is that security is getting woven into the agent platform narrative rather than treated as an add-on.
Gemini Enterprise Agent Platform
The keynote announcements Kurian previewed in the interview include Gemini Enterprise Agent Platform, described as the evolution of Vertex AI into a platform for building, governing, and optimizing agents at scale. Related launches included Agent Studio, Workspace Intelligence as a semantic layer across Google’s productivity suite, and Gemini Embedding 2. The agent platform exposes 200+ models via Model Garden and supports Google’s current model stack.
Thompson describes the interview as a preview of the keynote rather than analysis of the releases themselves.