Key Innovations Announced at Google Cloud Next

Vertex AI is no more; Gemini Enterprise is what comes next. For the past few years each spring has brought sweeping changes to Google Cloud's approach to enterprise AI. 2024 introduced agents and Vertex AI Agent Builder. 2025 was a code-first rebuild of the platform around the Agent Development Kit (ADK), Agent Engine, and Agent Garden. This year felt different. The branding changes were as sweeping as ever, but architecturally, this year was about depth and refinement. The center of gravity moved further down the stack into the unglamorous but essential work of supporting the production agentic workloads. Four things stood out: 🔹 Vertex AI is giving way to the new Gemini Enterprise Agent Platform. The new structure: Gemini Enterprise is the umbrella product, Agent Platform is the developer surface, the Gemini Enterprise App is the knowledge-worker surface, and the Marketplace is where partners plug in. Agents are now the organizing concept for Google Cloud's AI portfolio, a bet that the enterprise agent market is larger than the MLOps market Vertex was built for. One catch: the Gemini and Vertex AI APIs will reportedly coexist indefinitely, so the brand unification once again runs ahead of the developer experience. 🔹 This year's new products focus on agent governance and optimization, vs building and scaling. Agent Identity, Registry, Gateway, Anomaly Detection, Simulation, Evaluation, Observability, Optimizer were all rolled out at Next. This is the operational layer the platform was missing. 🔹 The 8th-generation TPU is a literal doubling-down on custom silicon and a substantial leap in performance. Google has split the TPU line into two chips, the 8t (training) and 8i (inference). The 8i claims 80% better performance-per-dollar over Ironwood, 3x on-chip SRAM, and a new acceleration engine that cuts on-chip latency by up to 5x. The architectural choices line up cleanly with the high-concurrency, long-context patterns of today's reasoning models and agentic inference. 🔹 The announcements actually shipped. In prior years, much of what was announced was "coming soon," sometimes for months. This year I logged into Agent Platform during Next and it was (mostly) all there. The DX, docs, and console were in-step with the keynote in a way I haven't seen Google Cloud execute in years. The platform is more complete and coherent than anything Google or its peers had on offer just prior to Next. The open question is whether developer education and advocacy can keep pace with a surface area that has grown enormous: ADK, Agent Studio, Runtime, Memory Bank, Sandbox, Sessions, Registry, Gateway, Identity, Simulation, Evaluation, Observability, Optimizer, Model Armor, A2A, MCP, AP2, A2UI… Full breakdown in the comments, including more Agent Platform takeaways and infra news, and my chats with Jeff Dean and Riyaz Habibbhai. If you're building agents, where are you most stuck right now: Build, evals, ops, or elsewhere?

I’m back meeting with customers after #GoogleCloudNext, and here’s what I’m hearing resonate the most - Agentic AI: Agents are top of mind across customers, and are poised to play an increasingly vital role throughout organizations. They are excited about our new capabilities - from Agent Development Kit to Agent2Agent Protocol to Agentspace enhancements. Infrastructure: We talked a lot about Ironwood, our most powerful TPU to date, but storage is also a critical component for minimizing bottlenecks in both training and inference. We introduced three new innovations that improve latency between 70%-500% - Hyperdisk Exapools, Anywhere Cache and Rapid Storage. All of these AI Hypercomputer, hardware and software enhancements together enable us to deliver more intelligence at a consistently low price. Gemini 2.5 models: There’s lots of excitement around Gemini 2.5 Flash, and developers can now start building with Gemini 2.5 Flash in preview on Vertex AI. This is our most cost effective model with "thinking” built-in. It adjusts the depth of reasoning based on a prompt’s complexity - and customers can control the performance based on their budgets. https://lnkd.in/gk_aeFkT

Google Cloud Next: Key Insights for AI Devs 🚀 Just wrapped up an inspiring Google Cloud Next, and wanted to share the highlights that I think are particularly relevant for those of us building the future of AI. A major takeaway was the focus on infrastructure built for the next wave of AI. 👉The new TPU v7 "Ironwood" is a beast, offering the power and memory bandwidth needed for the increasingly complex models we're working with. This isn't just about training; it's about having the horsepower to continuously run sophisticated AI. What really stood out to me was Google's strong push into making agent development a reality. This shift is huge for how we'll be building AI going forward. Key elements for developers include: 🟢 Agent2Agent (A2A) Protocol: This shared language will be crucial for building systems where different AI agents can communicate and collaborate effectively across models and tools. 🟢 Vertex AI Agent Builder: This new tool looks incredibly promising for streamlining the process of creating agents with integrated tools, memory, and reasoning capabilities. 🟢 Gemini Code Assist: Having more powerful AI-powered copilots directly integrated into the development workflow will be a game-changer for productivity. It's clear that Vertex AI is evolving into a comprehensive platform designed specifically for building and deploying these intelligent agents – going beyond just model training. We're seeing a move towards thinking in terms of context management, tool orchestration, and understanding the long-term behavior of AI systems. Ultimately, the future of AI development is pointing towards building coordinated, persistent systems that can learn, plan, and interact with their environment in real-time. This means focusing on things like long-term memory, multi-step decision-making, and seamless integration with various tools and other agents. Link to a more detailed overview in the comments Richard Seroter Karl Weinmeister Jeff Dean Thomas Kurian Oriol Vinyals Ivan 🥁 Nardini (Another highlight from the week was @arizeAI being announced in the keynote!)

Google Cloud Next 2025 marks a clear step forward in the evolution of agentic AI. This year’s announcements moved beyond foundational models and single-use copilots to a more structured system-level approach to AI deployment, focused on coordination, modularity, and scale. The most important developments that Google introduced: 1. 𝐀𝐠𝐞𝐧𝐭 𝐃𝐞𝐯𝐞𝐥𝐨𝐩𝐦𝐞𝐧𝐭 𝐊𝐢𝐭 (𝐀𝐃𝐊): A modular framework for building agents with memory, tool use, goal tracking, and long-term reasoning. It lets developers define planners, executors, and memory modules explicitly, rather than relying on prompt engineering alone. 2. 𝐀𝐠𝐞𝐧𝐭-𝐭𝐨-𝐀𝐠𝐞𝐧𝐭 𝐏𝐫𝐨𝐭𝐨𝐜𝐨𝐥 (𝐀2𝐀): A vendor-neutral standard that allows agents to communicate securely across platforms. This fills a key gap in interoperability, enabling organizations to deploy systems where specialized agents collaborate on complex workflows. 3. 𝐕𝐞𝐫𝐭𝐞𝐱 𝐀𝐈 𝐀𝐠𝐞𝐧𝐭 𝐄𝐧𝐠𝐢𝐧𝐞: A managed runtime environment for deploying and scaling agents in production. It includes lifecycle management, observability, and autoscaling support, helping teams move from experimentation to deployment. These were backed by advances in the underlying stack. Gemini 2.5 Pro showed marked improvements in reasoning, and the new Gemini Flash variant is optimized for speed and cost. Ironwood TPUs offer a tenfold increase in inference throughput. The integration of vLLM on TPU now supports high-throughput, low-latency model serving. BigQuery added native LLM functions and SQL-based access to generative models, bridging the gap between data infrastructure and GenAI. The rate of progress is notable. In 2023, the focus was on foundational models, basic tool use, and extensions. Agents were not part of the conversation yet. PaLM 2 and RLHF were early signals of fine-tuned control, but orchestration and memory were limited. In 2024, Gemini 1.5 introduced long-context windows and Agent Builder brought low-code agents into reach, though still within narrow scopes. By 2025, the platform has grown into a full ecosystem for agentic systems, complete with reusable components, shared protocols, and enterprise-ready runtime support. Within two years, the platform matured from single-assistant tools to something resembling an operating system for autonomous systems. Google’s contribution this year is a serious step toward practical deployment of agentic systems. What it does not yet solve are the layers of intelligence that come from learning, self-correction, and strategic reasoning. These are the areas where collaboration between research and industry will be most critical going forward. It was exciting to see McKinsey (Pallav Jain) present and part of the conversation at Next 2025. Because as much as this is a technical leap, it is not a technology-only challenge. Technical maturity without organizational readiness is no better than capability without application. Scaling agentic AI across the enterprise will require both.

Well, today #GoogleCloudNext has been quite the event with a huge number of announcements coming out of Las Vegas. From my vantage point, it’s interesting to see Google shift its messaging from building data estates to building systems of intelligence, systems that can lead to system of action with a little help from Agentic AI. At a high level, it seems to me they are at last elevating their data portfolio to actually stand next to their AI capabilities (primarily Gemini and Gemma model families) in terms of innovation, performance, and manageability. In other words, Google is catching up with both itself and its main rivals. Within the realm of data intelligence, analytics, and infrastructure, they’re doing this by trying to solve three distinct problems: 1. Breaking down data silos. They are tackling this via the new Knowledge Catalog, which has evolved from Dataplex to serve as a dynamic context engine. Paired with their new Cross-Cloud Lakehouse (which is standardized on Apache Iceberg), they are enabling zero-copy integrations with external clouds (AWS now and Azure later this year) as well as platforms like ServiceNow, SAP, and Workday — allowing query access without the friction of data movement. Basically, so long as data sits in Apache Iceberg, Google can seamlessly query that data. 2. Surfacing dark data. To make sense of semi- and unstructured data, Google introduced Smart Storage natively into Google Cloud Storage. As soon as files land in their storage layer, this feature automatically tags and embeds them. When paired with the Knowledge Catalog and Gemini, this service actively extracts entities and maps complex business relationships, creating a dynamic map of how data actually relates to the business — a major win for agentic processes starved for both meaning and relationships within the data they’re reasoning over. 3. Replacing manual coding with agentic tooling. Google is actively pushing to replace manual pipelines with intent-driven engineering. For data practitioners, the new Data Agent Kit drops directly into environments like VS Code to autonomously orchestrate Python, Spark, and SQL. Meanwhile, business users are now getting Conversational Analytics to chat with live data, data that’s not just sitting in a data lakehouse but data that’s typically trapped in on-device spreadsheets. All of this sits within a highly unified control plane via the Gemini Enterprise Agent Platform. Fascinating times here, to be sure, with Google investing so heavily in building a unified data layer and agentic control plane. It’s tempting to say that Google is simply trying to “own” the entire stack. Do they need to displace their competitors and partners to win here? Absolutely not. I think they will do well if they can create what I like to think of as distinct pockets of gravity, where it simply makes the most sense to run workloads within the Google Cloud ecosystem. The Futurum Group #AI #Data Google

Google Agent Development Kit (ADK) - yet another agent framework (YAAF 🤔) ? I know what you're thinking - does the world really need another agent framework? Well, let me try to make a case for it 🙃 First off, please check out the demo below from Franziska Hinkelmann, Ph.D. from Google's Cloud Next Developer Keynote. She builds a pretty impressive proposal agent live on stage, and in doing so, really shows off some of ADK's highlight features: ⚙️ 𝙎𝙩𝙖𝙣𝙙𝙖𝙧𝙙𝙞𝙨𝙚𝙙 𝙏𝙤𝙤𝙡 𝙄𝙣𝙩𝙚𝙜𝙧𝙖𝙩𝙞𝙤𝙣 In the demo, Fran defines an agent using LLMAgent, letting Gemini handle the reasoning. But ADK doesn't lock you into just LLM-driven decisions. It also provides WorkflowAgents (for strict sequences, loops, parallel tasks) and CustomAgents where you write the exact Python orchestration logic. This is key because it lets you blend deterministic control (vital for reliable business processes) with AI flexibility, offering a level of precision that's sometimes harder to guarantee when only prompting an LLM for complex flows. 🔌 𝙏𝙤𝙤𝙡 𝙪𝙨𝙚 In the demo Fran highlights the ADK's built-in support for Anthropic's Model Context Protocol (MCP). The agent uses MCP to perform RAG against a private database of building codes via Google's open-source MCP Toolbox for Databases. This shows ADK integrating with standardised protocols for robust connections to our own data and APIs, alongside its built-in tools (like Google Search, Code Execution) and wrappers for tools from LangChain/CrewAI. ✨ 𝘿𝙚𝙫𝙚𝙡𝙤𝙥𝙚𝙧 𝙀𝙭𝙥𝙚𝙧𝙞𝙚𝙣𝙘𝙚 (𝘿𝙓) ADK comes with a neat local Dev UI which is more than just a chatbot – it can be used for testing, inspecting events, and handles multimodal input by uploading both text and a floor plan image. Behind the scenes, ADK provides integrated concepts like SessionService (handling conversation state/history) and ArtifactService (managing files, like the PDF the agent generated), giving you essential application plumbing out-of-the-box. ☁️ 𝘿𝙚𝙥𝙡𝙤𝙮𝙢𝙚𝙣𝙩 Fran's initial diagram shows ADK within Vertex AI Agent Engine. ADK is designed with production on Google Cloud in mind, offering easy deployment paths to managed services like Agent Engine or Cloud Run. But critically, it also provides the flexibility for standard Docker/Kubernetes support if you prefer self-deployment. If you're building complex agents where precision, integration, and a code-first approach matter, ADK offers some compelling advantages worth exploring.

The era of the “toy agent” is over. 🚀 Over the last two weeks, Google Cloud has quietly overhauled the entire agent ecosystem – from the Agent Development Kit (ADK) to the Vertex AI Agent Engine. If you’re an Architect or CTO, this is the moment where agents move from experimental to mission-critical. As someone helping enterprises build GenAI systems every day, these are the 3 shifts that really matter: 1️⃣ Operational maturity 🛠️ The new Vertex AI Agent Engine gives us deep observability: traces, metrics, and a User Simulator to safely test non-deterministic behavior.Pair that with the Go SDK and a one-command adk deploy, and we finally have a clean path from notebook → managed runtime → production. 🔥 The Engine is rolling out to Australia and SE Asia by end of Q4, solving for latency-sensitive workloads. 2️⃣ Self-healing logic 🧠 The updated ADK introduces “self-healing” tool use. Agents can detect when a tool call fails and automatically retry with new parameters or strategies. That’s how you move from “it works in 80% of cases” to something you can trust at scale. 3️⃣ Enterprise governance 🔐 The real game-changer: agents now have first-class IAM identities. Combined with Model Armor, Security Command Center integration, Private VPC, and a generous free tier for Agent Engine, security and cost controls are baked into the platform – not bolted on later. 📚 Want to go deeper? Google just released a 54-page “Introduction to Agents” whitepaper and a free 5-Day AI Agents Intensive with Kaggle . We’re moving fast from “chatbots on documents” to governed, intelligent systems that behave more like microservices. 👉 In your 2026 roadmap, are agents still side projects – or part of your production architecture? #GoogleCloud #VertexAI #AIAgents #GenAI #SoftwareArchitecture #DevOps #CIO #CTO

Google’s Firebase Studio, unveiled at Google Cloud Next 2025, isn’t just another low-code tool—it’s a bold reimagining of how we build apps, and at least at first glance, looks to be fulfilling the promise of low-code app development. No install, cloud-native environment where developers can prototype full-stack applications using natural language prompts (“create a dashboard with real-time analytics”), tweak them conversationally, and deploy them with Firebase and Google Cloud integration baked in. It’s AI-powered development that bridges no-code simplicity with the flexibility and control devs crave from low-code platforms. This could sidestep the steep learning curves and ecosystem lock-in that platforms like Mendix, OutSystems, and Microsoft Power Platform often demand. The implications? Huge. By lowering the entry barrier—without dumbing things down—it could draw in startups, solo devs, and even business folks with big ideas but little coding know-how. Typical free tier capabilities with easy pay-as-you-go pricing make it a game changer. I’ve logged enough time in the dev trenches to know speed is king, and this could slash prototyping time from days to hours. The big question: will it open up dev work to more people than tools like GitHub Copilot or Cursor? I think it will, thanks to its lower barrier to entry—natural language prompts make it feel less like coding and more like collaborating with an AI. It’s untested at scale so far, and I’ll be digging deeper with full-cycle dev projects. More to come. What’s your take—game-changer, gradual evolution, or just another flash in the pan? #FirebaseStudio #LowCode #AIinDev

Today, at #GoogleCloudNext, we’re announcing significant improvements to Google Kubernetes Engine (GKE) to help platform teams succeed with AI: * Cluster Director for GKE, now generally available, lets you deploy and manage large clusters of accelerated VMs with compute, storage, and networking — all operating as a single unit. * GKE Inference Quickstart, now in public preview, which simplifies the selection of infrastructure and deployment of AI models, while delivering benchmarked performance characteristics. * GKE Inference Gateway, now in public preview, provides intelligent routing and load balancing for AI inference on GKE. * A new container-optimized compute platform is rolling out on GKE Autopilot today, and in Q3, Autopilot’s compute platform will be made available to standard GKE clusters. * Gemini Cloud Assist Investigations, now in private preview, helps with GKE troubleshooting, decreasing the time it takes to understand the root cause and resolve issues. * With a built-in partnership with Anyscale, RayTurbo on GKE will launch later this year to deliver superior GPU/TPU performance, rapid cluster startup, and robust autoscaling. More details in the blog post below... https://lnkd.in/gxuzMJaJ

Google Cloud Next '25 left no doubt: we're entering the Agentic AI era. Here are some key announcements fueling this shift: 1️⃣ Agent Development Kit (ADK): Building Made Easier Google launched the Agent Development Kit (ADK), a powerful open-source framework designed to simplify the entire lifecycle of building and deploying sophisticated agents and multi-agent systems. Based on Google's internal experience, ADK offers developers greater flexibility and precise control, enabling the creation of production-ready, scalable agentic applications. Think easier development, multi-agent coordination, built-in streaming capabilities, and robust evaluation tools. 2️⃣ Agent2Agent (A2A) Protocol: Enabling Collaboration How do agents built on different platforms work together? With the new Agent2Agent (A2A) protocol! This open standard, developed with contributions from over 50 industry leaders, provides a common language for AI agents to securely communicate, exchange information, and coordinate actions across various enterprise applications and services, regardless of the framework or vendor. This is crucial for creating truly interconnected and collaborative AI ecosystems. 3️⃣ Agentspace Enhancements: Discovery & Adoption Google's Agentspace, the platform connecting work apps with Google Search and AI agents, received updates making it easier for organizations to discover, create, and adopt agents. This includes a new no-code Agent Designer and easier access to pre-built agents, simplifying how employees can find information across the business, synthesize insights, and automate tasks securely.