Welcome to the August 1–15, 2025 edition of Google Cloud Platform Technology Nuggets. The nuggets are also available on YouTube.
Google has announced a Developer toolkit for scaling Agent2Agent (A2A) agents on its platform. The toolkit includes:
- version 0.3 of the A2A protocol, that introduces gRPC support, security card signing, and extended client-side support in the Python SDK.
- Native support for A2A in the open-source Agent Development Kit (ADK).
- Developers have flexible deployment options for A2A agents, including Agent Engine, Cloud Run, and Google Kubernetes Engine (GKE).
- Agentspace is being introduced as a destination for A2A agents to meet end users, providing critical governance, safety, and control features.
- The Vertex GenAI Evaluation Service now supports A2A agent evaluations.
Check out the blog post for more details.
Gemini Code Assist is now integrated directly into your GitHub workflow at no charge. When a pull request is created, Gemini is automatically assigned as a reviewer and gets to work immediately, providing pull request summaries, in-depth automated reviews and more. Check out the blog post for more details.
Colab Enterprise notebook has got new features that include Data Science Agent (DSA) for automating end-to-end data science tasks, multi-cell code generation for diverse programming needs, easy visualisation creation from simple prompts, and automated error explanation and fixing. Check out the blog post.
Looking to understand a complex Agent put together with Agent Development Kit (ADK). Check out this deep dive (with code) blog post that dives into building a research agent for lead generation. The post highlights how to structure a complex task into a hierarchy of cooperative agents, manage state across interactions, and design for parallelism to create a system that is both powerful and efficient.
OpenAI announced gpt-oss models (gpt-oss-120b and gpt-oss-20b and Google Cloud has made them available on Google Kubernetes Engine (GKE). It is available via the GKE Inference Quickstart, that provides validated, performance-tuned deployment recipes that let you serve state-of-the-art models with just a few clicks. Check out the blog post for more details.
Container image streaming in GKE has helped to improve the image pull times and accelerate application startup. This feature has some significant enhancements, which “can help your GKE workloads start up faster and run more efficiently, particularly ones suffering from long startup times due to large container images. Specifically, AI/ML model serving applications will benefit from the improved startup times”. A key improvement is the new intelligent read-ahead capabilities. These allow GKE to proactively fetch image data that is likely to be requested next, minimizing the time your applications spend waiting for data during startup. Check out the blog post for more details.
Gartner has recognized Google as a Leader for the third year in a row in the 2025 Gartner® Magic Quadrant™ for Container Management. Check out the blog post and download your complimentary copy of the 2025 Gartner® Magic Quadrant™ for Container Management.
GKE has announced a preview of multi-subnet support for clusters. This enhancement removes single-subnet limitations, increasing scalability, optimizing resource utilization, and enhancing flexibility of your GKE clusters. This functionality is supported for all clusters, using GKE version 1.30.3-gke.1211000 or greater. Check out the post.
What is a dangling bucket attack? I will quote from the blog post, “It happens when you delete a storage bucket, but references to it still exist in your application code, mobile apps, and public documentation. An attacker can then simply claim the same bucket name in their own project, effectively hijacking your old address to potentially serve malware and steal data from users who unknowingly still rely on a bucket that is no longer officially in use.” If this got you concerned, check out the best practices to prevent dangling bucket takeovers.
The first Cloud CISO Perspectives for August 2025 is out. It highlights key findings from Google Cloud’s Office of the CISO’s “Cloud Threat Horizons report”. It focuses on foundational vulnerabilities like credential compromise and misconfiguration, the increasing trend of targeting backup infrastructure, and the evolving methods for bypassing multi-factor authentication.
If you are looking for a quick overview of what Google Cloud has announced in the Data areas : Database, Analytics and Business Intelligence, then check out and bookmark this link that has the latest updates.
Google’s Data Cloud, a unified, AI-native cloud platform has fully adopted the agentic shift that we are seeing. Its focused on 3 main areas: a new suite of specialised AI data agents, an interconnected network for agent collaboration, and a unified, AI-native data foundation. Key thing to highlight here are the new agents that include:
- Data Engineering Agent (Preview) in BigQuery for automating complex data pipelines.
- Data Science Agent (Preview) in Colab Enterprise Notebook for autonomous analytical workflows.
- Conversational Analytics Agent with Code Interpreter (Preview) for advanced business analysis using natural language.
Included in this update is the Looker MCP integration that we talk about next along with Spanner Columnar Engine integration that we discuss a bit later in this post. Check out the post for more details to understand the new ways that you can interact and get insights across your Data cloud.
MCP Toolbox for Databases is an open source MCP server for databases. It supports various databases in Google Cloud. Looker has now integrated their MCP server into this toolbox and that makes the task of integrated Looker into various AI clients much easier. You can now ask natural language queries in AI clients like Gemini CLI, Claude Code, etc to interact with Looker. Check out the blog post for more details.
Google Cloud’s Spanner columnar engine, is a new feature designed to unify online transaction processing (OLTP) and analytical query processing (OLAP) within a single database. The engine achieves this integration by combining columnar storage, which optimises for analytical queries by reading only relevant data, with vectorised query execution, which processes data in batches for improved CPU utilisation. Some of the benchmarks are interesting. Analytical queries on live operational data are said to have been boosted upto 200 times. Check out the blog post for more details.
Enhanced Backups for Cloud SQL is a new service designed to increase data protection for Cloud SQL instances. It does that by providing immutable, logically air-gapped backup vaults that are separate from the source project to defend against various threats. Check out the blog post for more details and getting started.
If you are looking to understand how Google Cloud goes about Platform Engineering, there are some key points that have been identified in this blog post, which is based on a recent talk at PlatformCon 2025. The talk focuses on the “shift down” approach “that advocates for embedding decisions and responsibilities into underlying internal developer platforms (IDPs), thereby reducing the operational burden on developers.” Check out the blog post for more details and specifically key learnings in their journey.
“Moving from a trained model in a lab to a scalable, cost-effective, and production-grade inference service is a significant engineering challenge. It requires deep expertise in infrastructure, networking, security, and all of the Ops (MLOps, LLMOps, DevOps, etc.)”. To help enable that, Google Cloud has announced the GKE inference reference architecture: a comprehensive, production-ready blueprint for deploying your inference workloads on Google Kubernetes Engine (GKE). Check out the article and the companion repository.
If you are using Agent Development Kit (ADK) for creating conversational AI agents, you should check out this deep dive article that explains how ADK manages agent state and memory to enable personalised user experiences. The article illustrates these concepts through a Python Tutor agent, explaining short-term memory for in-session recall and long-term memory for persistent learning history, including storage options like SQL databases and Vertex AI Memory Bank.
If you have been deploying your AI applications on Google Cloud, one of the challenges has been to come up with an effective strategy to host your model artifacts. As the article states, “Baking models into container images leads to slow, monolithic deployments, and downloading them at startup introduces significant delays”. The article suggests that the better way is to decouple your model from the application by hosting them in Cloud Storage and accessing them efficiently from GKE and Cloud Run. But storing them simply in Cloud Storage is not enough. You need to have efficient strategies to load these models. Check out the blog post for more details.
A term that I frequently use when evaluating any developer technology is “Time to First Hello World”. Ideally it should be with as minimal a friction as possible and with good quickstart guides and samples. Google Cloud Developer Experience team’s goal is simple: “to help developers get from learning to launching as quickly and effectively as possible.” And as the blog post states, hands-on documentation and the ready-to-use code samples are the key drivers to making that experience possible. But given the rate at which things are changing, it is challenging to keep the documentation and the samples updated. There is nothing more frustrating that downloading a library and finding that the documentation is not upto speed or there are not enough relevant code samples to try out the library or the new feature introduced. The team has been piloting the use of Generative AI, specifically using Gemini, to help in this objective. Check out the blog post that gives a glimpse into how AI is being employed to ensure delightful developer experiences.
The Gemini Live Multimodal API allows developers to stream data, such as video and audio, to a generative AI model and receive responses in real-time. If you are looking to understand a real world example of how to put together a scenario using the Live Multimodal API, check out this blog post, that covers how to build an automated quality inspection system. The system described uses a live camera feed to analyze products. In real-time, it first identifies products using barcodes or QR Code, and then detects, classifies, and measures visual defects simultaneously. This along with responds for defects and alerts for severe issues. Check it out.
One of the challenges that you face while doing any Cloud deployments, Google Cloud or otherwise, is to ensure that you are following best practices when it comes to deploying and configuring the services correctly, along with security settings, observability and monitoring and more. While there are checklists that have been available for a while, Google Cloud now provides Cloud Setup directly in the console to help you with steps to deployment your workload along with a Terraform script to automate the whole process, depending upon some of your preferences. The Cloud Setup provides you with 3 types of options: Proof of concept, Production and Enhanced Security. Depending on your choices, it then constructs the right kind of setup for you. Check out the blog post for more details. Cloud Hub Optimization and Cost Explorer are now in public preview, and they help you get the necessary insights. These tools also support the application centricity that was introduced via App Hub. Check out the post for the kinds of reports that are now available.
If you are responsible for running your Cloud Operations and specifically when it comes to cost, you typically have to grapple with queries like which resources have been the most expensive, which resources have been over-provisioned, etc.
Gartner® has named Google a Leader in the Gartner Magic Quadrant™ for Strategic Cloud Platform Services. Google is now positioned furthest for completeness of vision. Check out the blog post and download your complimentary report here: 2025 Magic Quadrant for Strategic Cloud Platform Services.
Q2 2025 AI Hypercomputer updates are here. A few of the updates include:
- New flexible consumption models are available via Dynamic Workload Scheduler, including Calendar mode for assured capacity and Flex start mode for better economics on-demand.
- Cluster Director has gained new capabilities, including a new GUI, observability, and straggler detection features, to simplify large-scale cluster management. If you are curious about straggler detection features, check out this fantastic dive into what stragglers are, the domino effect that they can have on AI infrastructure and how its being addressed in future, when it comes to infrastructure.
- A new monitoring library for Google Cloud TPUs has been released, providing granular insights into performance and accelerator utilisation.
If you are looking to learn more about designing and building Agentic apps on Google Cloud and if you live in America, you should check out this 1-day in-person workshop that helps you experience a hands-on engagement to participate in a gamified mission where you’ll learn an end-to-end roadmap for taking an AI idea from its initial concept to full-scale operational reality. Check out the blog post to see if this event is happening in your city and register soon.
This is another in-person developer workshop series that is coming to your city. It is a hands-on event where with a deep dive into code, deployment, and the production-grade patterns that turn AI ideas into real-world solutions. The focus is on using Cloud Run as the platform to host your AI applications. You get to do hands-on labs that include building a MCP Server and deploying it on Cloud Run, Building an Agent using Agent Development Kit, containerize it with Docker, deploy it to a Cloud Run instance with GPU acceleration. Dates are out for multiple US cities and you should register for it. EMEA, APAC dates will be out soon. Check out the blog post for more details.
If you would like to share your Google Cloud expertise with your fellow practitioners, consider becoming an author for Google Cloud Medium publication. Reach out to me via comments and/or fill out this form and I’ll be happy to add you as a writer.
Have questions, comments, or other feedback on this newsletter? Please send Feedback.
If any of your peers are interested in receiving this newsletter, send them the Subscribe link.
Source Credit: https://medium.com/google-cloud/google-cloud-platform-technology-nuggets-august-1-15-2025-c1b054104647?source=rss—-e52cf94d98af—4
