Author: Richard Seroter

Today’s presentation at the Major League Baseball event was super fun. It was also good to debate and discuss what sort of software tactics and AI strategies that teams of 5-20 engineers should embrace right now. Lots of opinions!

[article] How to Manage Context in AI Coding. It’s not just about cleverly-written prompts. How do we think about context? This post has some good advice for doing it well.

[article] Developers don’t care about Kubernetes clusters. That’s true. But they’re forced to when things go wrong even if the app is “correct.” I wonder if it’ll be less about abstractions and overlays, and more about using AI to troubleshoot when you don’t know all the k8s commands.

[blog] GKE: From containers to agents, the unified platform for every modern workload. With Kubecon going on this week, you’ll probably see a lot of Kubernetes. This post from Drew is outstanding and shows all the progress being made.

[blog] Introducing Agent Sandbox: Strong guardrails for agentic AI on Kubernetes and GKE. This is a smart capability that’s now part of the core Kubernetes engine. It should reduce the risk/concern people have about agents running wild.

[blog] Boosting LLM Performance with Tiered KV Cache on Google Kubernetes Engine. These are some pretty wild performance improvements for those doing LLM inference on Kubernetes.

[article] Microsoft’s .NET 10 arrives with AI, runtime, and language improvements. Feels like a lowkey announcement set. I know I don’t watch this ecosystem as closely as before, but I haven’t noticed a ton of chatter about these major updates.

[blog] Integrating File Search with the Gemini CLI Extension. We just shipped this RAG-as-a-service feature for the Gemini API, and someone already built a cool open extension for the Gemini CLI. Amazing.

[blog] GitHub in 2025. Fascinating business unit to watch. Stephen has a good summary of the change in pace and tone evident at the recent GitHub Universe event.

[blog] Building AI Agents with the GO Agent Development Kit (ADK). I think Bill wins the award for “first piece I’ve seen written about the ADK Go” from a non-Googler.

[blog] Accelerate debugging with Crashlytics MCP Tools. Why go visit all these dedicated surfaces to do work if you can bring the data to wherever you want with MCP? I think we’re seeing a shift happen in real-time.

[blog] 100% coverage is not that trivial. Do you need complete coverage, or better coverage? Mark talks about the genuine work it takes to have test coverage that matters.

Want to get this update sent to you every day? Subscribe to my RSS feed or subscribe via email below:

Was it a busy Monday for you? It was for me. Tomorrow I get to do a presentation at my favorite team’s ballpark as part of the MLB Tech Summit. So, that’s pretty awesome.

[blog] Predictions 2026: Software Development Goes From Jamming To A Full Orchestra. Diego has been close to this trend for a long time, so I listen to his perspective.

[blog] Build better software to build software better. Great post from the Slack team. If your build system is slow or painful, all the AI dev tools in the world won’t help you.

[blog] AlloyDB accelerates AI with automated vector indexing and embedding. This is kinda awesome. Auto embedding and auto indexing for your existing databases? That’s powerful.

[blog] Yes really, curl is still developed. Yes, this ubiquitous tool is still under active development. How? Why? Find out here.

[blog] Building AI Agents Visually with Google ADK Visual Agent Builder. Build some fairly sophisticated AI agents without slinging a lick of code? This new web builder included in our Agent Development Kit works pretty well.

[article] AI is all about inference now. Training was always going to be niche. Important, but not something everyone has to do. Inference is where it’s at. It’s good to have TPUs!

[article] Over one-third of companies plan to replace entry roles with AI, survey says. “Plan to” is different that “will do”, but nonetheless, we’re in an odd state.

[blog] Stop Vibe Coding Your Unit Tests. Real experience with these tools yields insights you won’t get from outside observers.

[blog] Go SDK Power-Up: ADK Support Arrives in MCP Toolbox. Build agents in Go, get access to a variety of data sources via MCP. Not a bad recipe.

[blog] The Go Ecosystem in 2025: Key Trends in Frameworks, Tools, and Developer Practices. Twice as many Go developers as five years ago? It’s a growing ecosystem, and here’s a look at what tools and tech matter.

[blog] Achieve better AI-powered code reviews using new memory capabilities on Gemini Code Assist. I like the idea of a code review agent that learns and adapts based on what your applied standards are. Check this out.

[blog] Terraform vs. Pulumi vs. Crossplane: Choosing the right IaC Tool for your platform. There are these, and I suspect another emerging paradigm for infrastructure management in the AI era.

Want to get this update sent to you every day? Subscribe to my RSS feed or subscribe via email below:

Check out a handful of AI-related things today, including app building advice, new capabilities, and an architectural view of an AI platform. Have a great weekend!

[blog] Introducing the File Search Tool in Gemini API. While we offer an amazing input context window, you don’t want to keep shoving tons of content into each request. With this new Gemini API feature, we just made RAG absolutely trivial.

[blog] Google’s Gemini Models on Databricks. You can use all sorts of models with Databricks, but I’m glad everyone now has access to the best one 🙂

[article] Survey: Where AI Reduces Toil and Where It Still Falls Short. There’s still a lot of toil out there, even when AI is available. Tool sprawl also seems out of control!

[article] Sam Altman says OpenAI has $20B ARR and about $1.4 trillion in data center commitments. I’m no mathematician, but doesn’t seem to exactly add up. But hey, smarter people than me are don’t seem too worried about it.

[blog] Your First AI Application is Easier Than You Think. Yes it is. This blog post refers to a codelab that you can follow along with to build your confidence as an AI app builder.

[blog] Announcing User Simulation in ADK Evaluation. I really like this functionality. It’s baked into the Agent Development Kit, and lets you define a high level goal and use the LLM to dynamically produce the user side of the conversation to pursue that goal.

[blog] Prototypes vs Products. Different goals and mindset for each. Marty clarifies here and tries to keep product managers from embarrassing themselves.

[blog] Announcing the Agent Development Kit for Go: Build Powerful AI Agents with Your Favorite Languages. Go devs are getting some great frameworks for building AI apps and AI agents. Check this one out.

[article] The Production Generative AI Stack: Architecture and Components. Good look at an AI/ML stack. Not necessarily an AI agent or app stack, but some overlap.

Want to get this update sent to you every day? Subscribe to my RSS feed or subscribe via email below:

Finished this trip to Google Cloud HQ with a productive day. My team and scope changed this week, so I’m focused now on landing all that while keeping us running at full speed. Easy, right?

[article] The Learning Loop and LLMs. Don’t skip the crucial stage of actually learning something, in favor of doing it faster.

[blog] Yes, we can have nice things! Using Gemini CLI in an Enterprise Environment. This is a much-needed post about the “responsible” way to use something like our agentic CLI.

[blog] A Reference Architecture for Composable Enterprise AI Systems. I saw this because of a tweet from Keith and his argument about what you get for ‘free” in a cloud versus DIY infrastructure. Orchestration (agentic and infra) can be taken for granted!

[blog] Announcing Ironwood TPUs General Availability and new Axion VMs to power the age of inference. Everyone seems to want to build their own chips now, and I’m happy we have such a head start. This new TPU is crazy powerful, and already depended on by key customers.

[blog] From silicon to softmax: Inside the Ironwood AI stack. Tons of details about our AI stack woven into this cutting-edge TPU.

[article] 21 Git Commands for Software Engineers. I’m somewhat glad that AI is allowing me to forget some of this, but we should still know what’s possible in Git (even if we forget the exact commands).

[blog] How I use AI (Oct 2025). I like these kinds of reports. This doesn’t feel too different from how I use AI things. I do like tagging git commits with “AI” if AI did most of the work. Good idea.

[blog] More ways to build, scale, and govern AI agents with Vertex AI Agent Builder. Our agent stack is legit. And now Go devs get an Agent Development Kit and our Agent Engine gets a free tier along with new observability experiences.

[blog] Architectural debt is not just technical debt. Where should enterprise architects focus and how does architectural debt propagate? Insightful post.

[blog] From Skeptic to Believer: A Process-Driven Approach to Vibe Coding. We’re collectively learning what works and what doesn’t with AI coding. Mollie shares her recent “aha” moments.

Want to get this update sent to you every day? Subscribe to my RSS feed or subscribe via email below:

Full day today, with another good set of reading list items that touch on AI coding, AI projects, and advances in AI technologies.

[blog] Conductors to Orchestrators: The Future of Agentic Coding. Important piece here. Maybe we all find our way to orchestration, or pull back to something else. But the attempt to initially get there will not stop.

[blog] Why I code as a CTO. What and why should leaders code? I liked this, as there is real value in maintaining a strong context, and doing what you love.

[blog] The Agentic Manifesto: Engineering in the Era of Autonomy. Casey outlines the new rules for engineering leaders who are going to build with AI.

[blog] Quick Guide to ADK Callbacks. The ability to intercept and intervene before/after an agent, model, or tool does something is powerful.

[blog] Is it worrying that 95% of AI enterprise projects fail? I find that the most important question to ask someone when the raise an issue of cost, success, failure, complexity, or whatever is “compared to what?” That’s a question Sean asks here.

[blog] Driving a web browser with Gemini’s Computer Use model in Java. Guillaume’s experience wasn’t perfect, but he shows the potential of working with AI that can programmatically interact with your computer.

[article] A fresh look at the Spring Framework. Speaking of Java, Spring has remained relevant and modernized along the way.

[article] Does Giving Developers Control Over AI Improve Trust and Adoption? Hmm. This seems like very actionable data if you’re trying to figure out how to establish sustainable adoption of AI tech at your business.

[blog] Cloud CISO Perspectives: Recent advances in how threat actors use AI tools. Make yourself aware of how bad actors are using AI too.

[blog] The Smart Shopping Cart: AI Agents With Gemini, MongoDB Atlas Vector Search, and the MCP Toolbox. Good mashup of tech here that demonstrates how to use natural language to find similar data.

[blog] Code execution with MCP: Building more efficient agents. Don’t fall in love with too many AI patterns. They change too often.

[blog] Beyond Standard LLMs. Also, don’t fall in love with one type of model, as there’s constant innovation happening here too.

Want to get this update sent to you every day? Subscribe to my RSS feed or subscribe via email below:

Great day at the office, and a terrific reading list (if I may say so) today. Many good “think” pieces versus product announcements.

[blog] Most of What We Call Progress. Wonderful post on the things that many of us tend to learn over time while building systems.

[blog] Leaving the Cloud Isn’t for Everyone. Right on. For a subset of folks, running their own hardware and systems is the right thing competitively, and from a business sense. For the other 93%, you’re wasting time and money doing it yourself.

[article] Generative AI in the Real World: Chris Butler on GenAI in Product Management. Good discussion (and transcript for those who read faster than they listen). Product management is ripe for taking advantage of AI.

[blog] What have we learned about building agentic AI tools? I agree with most of these, but not sure RAG-for-codebases being completely wrong. Do repo search tools replace that? Not sure.

[blog] The Data Engineering Agent is now in preview. Looks cool. It doesn’t look like it replaces a data engineer, as you still need to know what to ask for. But this type of assistance makes smart people better.

[article] What Good Software Supply Chain Security Looks Like. Not a “hot” topic like it was a few years ago, but no less important.

[blog] Why Log Data Management Is a Thing. It’s a question of data, says James. Here’s a good roundup of some of the players trying to get log management under control.

[blog] Medium CEO explains how AI is changing writing. Writing to learn and think is important. It’s why I don’t use any AI on any of this, or other newsletters I write. But other circumstances don’t require that so much “thinking” to occur.

[blog] How Design Teams Are Reacting to 10x Developer Productivity from AI. Good timing here, as I’m talking to a group of designers today about the expected impact of AI.

[blog] Accelerating the magic cycle of research breakthroughs and real-world applications. We’re going faster and discovering more things in Google Research because of AI. Great to see.

Want to get this update sent to you every day? Subscribe to my RSS feed or subscribe via email below:

I’m in Sunnyvale this week for four different Google offsites. It’s a key time for ensuring momentum through 2025, and identifying our biggest (initial) priorities for 2026.

[report] State of Developer Ecosystem Report 2025. Some excellent data in this survey of 24k+ developers. What languages have momentum, where are developers using AI, and more.

[article] Google ADK: A complete toolkit for production-ready AI agents. It’s a great framework for those that want baked-in opinions when building AI agents.

[blog] How fast can an LLM go? Maybe most appealing to ML engineers, this post was still interesting to me.

[blog] The real problem with AI coding. Are we piling up “comprehension debt” into our repos as we stuff in code that wasn’t written by humans? How do you avoid it?

[blog] Beyond Request-Response: Architecting Real-time Bidirectional Streaming Multi-agent System. This post presents a sophisticated topic, but basically, it explains why the old “take-turns” method for AI agents is too slow. It describes a new, faster way that allows agents to process data in real-time, much like a continuous, natural conversation.

[article] Stop Writing Code, Start Writing Docs. My boss Keith has a long history with dev tools, and continues to use them on a daily basis. Here’s some perspective worth listening to (or reading).

[blog] Evolving Ray and Kubernetes together for the future of distributed AI and ML. These projects are working better together, which is good news for AI/ML folks who depend on Kubernetes. Also, great new TPU support.

[blog] The Green Tea Garbage Collector. Do you geek out over low level functions like garbage collection? If so, you sound interesting. Let’s be friends.

[blog] The New Calculus of AI-based Coding. Good thoughts here about a productive approach to AI coding, but with cautions about what breaks at this new velocity.

[blog] New tools in Google AI Studio to explore, debug and share logs. Better insights and tracking capabilities are baked into this must-have Ai tool.

Want to get this update sent to you every day? Subscribe to my RSS feed or subscribe via email below:

That’s a wrap on the week, and the month. Tonight I’ll chase my kids around the neighborhood for Halloween, and “borrow” some of the top-tier candy they acquire. Don’t judge me.

[blog] Spec-Driven Development: 10 things you need to know about specs. This post contains some very good questions to consider. It’ll be easy for us to over-index on specs and accidentally re-create waterfall development!

[blog] What do we think of GitHub saying there are 180m developers in the world? It’s fine. They’re counting anyone with a GH account which is generous. But the macro point is that there are more builders out there, using all sorts of tools.

[blog] Analyze anything with AI-powered SQL in BigQuery. I really love this feature. So much that I tried it out today and did a tweetstorm about it.

[blog] Build a Retail Agent with MCP Toolbox, YugabyteDB, and Google ADK. You can fulfill one of these sorts of sneaky-powerful use cases in minutes and hours nowadays. Wild.

[blog] StreetReaderAI: Towards making street view accessible via context-aware multimodal AI. We’ll keep doing commercial stuff to pay the bills, but here’s another great case of AI research that might have a massive impact on those who are blind or vision impaired.

[blog] Google Cloud Assist Investigations: AI-Powered Root Cause Analysis. “I don’t know where to start” is going to be a question we ask ourselves less and less often. Ops team RIGHT NOW can get an AI assist before they even start troubleshooting.

[blog] Survey: Two-thirds of AI-native startups let AI write most of their code. I’d assume these shops will need that AI-assisted troubleshooting if they run these apps long enough for it to matter!

[blog] Tool descriptions are eating up all your AI tokens (but they don’t have to). Good points by Craig here. It’s super easy to eat up 100k tokens just on processing the available MCP tools.

Want to get this update sent to you every day? Subscribe to my RSS feed or subscribe via email below:

There’s a good mix of advice, industry news, and tech insight in the reading list today. Enjoy!

[article] The 4 Seasons Organizations Go Through—and How to Lead in Each. It took me too long to realize that leadership isn’t a blunt hammer that applies to every circumstance and person the same way.

[blog] Java relevance in the AI era – agent frameworks emerge. No shortage of frameworks for Java developers. Besides Python, it might have the best options for AI app/agent builders.

[article] Surviving the AI Sprint: up close with Google Cloud and Cisco. Fascinating discussion that shows how much innovation is happening in the infra tier, and why demand is outstripping supply.

[blog] Introducing ConfigHub. Maybe lost in all the AI stuff, there’s innovation happening up and down the stack. Brian and team are offering a new way to store configurations.

[article] What’s the Go language really good for? Most things, nowadays. But like every language, there are sweet spots and weaker areas.

[blog] Master multi-tasking with the Jules extension for Gemini CLI. Every developer is a “lead” now and coordinating the work of multiple parties.

[blog] How many pillars of observability can you fit on the head of a pin? Charity lays it down on people obsessed with pillars of observability, versus focus on unified storage model for signals.

[blog] Chat in NotebookLM: A powerful, goal-focused AI research partner. While wildly popular, NotebookLM is still underrated. And now it’s even more useful.

[article] GitHub Embraces the Coding Agent Competition With Agent HQ. Smart to drag agents closer to the source code in GitHub. We’ll see if that’s the true center of gravity moving forward!

[blog] Announcing docs.cloud.google.com: The new home for Google Cloud documentation. It’s not trivial moving 9m pages of docs and making a series of updates on the way. My team finished this first milestone, and it’s a great accomplishment.

[blog] Why do AI models use so many em-dashes? Nobody knows for sure, and I refuse to change how I write just because AI is stealing my style.

[blog] How I Made an MCP Server That Saves Me an Hour per Week. Aim lower. Not EVERY use of AI has to change the world. Making a particular task a little better is still a win.

Want to get this update sent to you every day? Subscribe to my RSS feed or subscribe via email below:

What a wild year or so of tech changes. Cloud provider preferences aren’t the same, toolchains aren’t the same, and well-established jobs (like product manager) aren’t the same. Today’s reading list exemplifies that.

[blog] Google Cloud Q3 revenue surges 34% as backlog hits $155 billion. The fundamentals look good, and customers are trusting us to help them do their next big thing.

[article] Amazon faces a ‘fundamental’ shift in how AI startups spend on the cloud, internal documents show. Behind a paywall, but if you have access, it’s a good read about the shifting preferences of startups who look at a new style of tech provider. Backed up by public data here.

[article] 7 Must-Know Agentic AI Design Patterns. If you’ve been keeping an eye on this space, these won’t feel brand new. But still a useful overview of some important ideas.

[blog] Introducing the Jules extension for Gemini CLI. This feels like a big deal. Offload background work to Jules and stay in the flow within Gemini CLI.

[blog] The Role of a Product Manager in the Age of AI. Building-first culture may spark the most disruptive cultural change within companies in a decade.

[blog] Introducing Spring AI Agents and Spring AI Bench. The Spring team never rests. This looks like an interesting pair of new projects for those using agentic coding tools.

[blog] Gemini CLI Tips & Tricks. Speaking of AI coding tools, you’ll be an absolute pro after you apply many of these expert-level tips from Addy.

[blog] The “10x” Commandments of Highly Effective Go. Good specific advice on the Jetbrains site. This is obvious for Go devs, but much of it applies to any programming language.

[blog] How Model Armor can help protect your AI apps from prompt injections and jailbreaks. This is a model-agnostic and cloud-agnostic service that offers a handful of very important dimensions of protection for your AI models and apps.

[blog] Introducing Cursor 2.0 and Composer. Big update for the AI coding tool with the most momentum. Folks love Cursor, and this won’t change that. More from Simon on their unique coding model.

[blog] Context engineering is sleeping on the humble hyperlink. Are we missing hyperlinks as an obvious source of context for your agents?

[blog] Comparing Agent Frameworks: PydanticAI, LangChain 1.0 and Google ADK. Genuine praise and criticism for popular agent frameworks, including ours.

Want to get this update sent to you every day? Subscribe to my RSS feed or subscribe via email below: