Author: Richard Seroter

Back in the saddle after a couple of days in Texas. Today’s reading list has some good hype-busting content about AI, and also some useful advice in a handful of areas.

[blog] Here’s which Google AI developer tool to use for each situation. New, from me. Here’s my take on how to choose from the handful of AI tools we offer. Let me know what you think of the heuristics here!

[blog] The problem with Claude Code and Cursor: The AI Coding “Death Spiral.” Yes, simple is good. Don’t tackle giant efforts at once. Deconstruct what you want from the AI tool, and don’t try to be so clever.

[article] 87% of game developers are using AI agents in their workflows, says new survey. I’ve noticed that the gaming industry is often quick to jump on new innovations. Here’s another example.

[blog] How much energy does Google’s AI use? We did the math. I’m fairly sure this data will surprise you. It surprised me.

[article] Managing AI coding assistants without losing control. Are you think of AI tools as shortcuts, or enhancers? Be intentional about what you want to use them for.

[article] How Reddit Delivers Notifications to Tens of Millions of Users. Always fun to learn how other people solve big tech challenges.

[blog] Build Day 0: Engineering the Virtual CTO Advisor with Google Cloud Vertex AI. Keith is building a “virtual advisor” and sharing what he’s been learning. Good pragmatic stuff. Same with this new post on prompt management.

[blog] Beyond guardrails: A taxonomy of platform engineering control mechanisms. Good post that helps us use the correct term for the situation.

[blog] Introducing ‘Gemini for Government’: Supporting the U.S. Government’s Transformation with AI. A ton of capabilities, at a bargain price. I’m glad to see governmental groups getting easier access to leading tech.

[article] The Hidden Cost of AI Pilots that Never Scale. Even quick pilots can accidentally have long-term costs. Think wisely about what investments you need to make, and what your “carrying costs” might be.

[article] GitHub’s Fall: Microsoft’s AI Takeover, Developer Betrayal, and the Next Fight for Digital Sovereignty. Yikes, quite a spicy take. But I do think this organizational change with GitHub and Microsoft has larger ramifications than most people think.

[blog] Build Better Agents in Java Than Python: Embabel vs Pydantic AI. Java devs have some of the best non-Python options for building agents.

[article] All the Biggest Announcements From This Year’s ‘Made by Google.’ The format yesterday was … interesting. I enjoyed parts of it, for sure. Here’s our summary of the news.

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

I felt my age this weekend. Busy days of sports and chasing kids left me WORN OUT. But, lots of fun and memories. I’m on vacation the next two days while I take a kid to college, so the reading list will be back on Thursday.

[article] Harsh truths to save you from ChatGPT psychosis. It’s definitely dangerous to treat AI like some sentient, all-knowing partner. That said, Forrest is wrong on points 1 and 4, partially right on 2 and 3.

[blog] 14 ways Googlers use AI to work smarter. How are we using AI within Google? Most of us are using it every day for different parts of our job. This offers a wide range of scenarios.

[blog] You Can Build Better AI Agents in Java Than Python. Rod teased it last week, and proved it today. Yes, this JVM-based agent framework looks simpler and more production-ready than a well-known Python option.

[blog] How Google does it: Your guide to platform engineering. This isn’t a product pitch, but more of a philosophical look at how to think about platform engineering.

[article] Diagnosing Your Company’s Strategy Problem. Figure out what’s going wrong with your strategy efforts and how to get back on track.

[blog] Scalable AI starts with storage: Guide to model artifact strategies. Great advice here in an area I haven’t seen much written about.

[blog] The Great SSL Certificate Panic. I didn’t know about these new fast-expiring certs. Kate gets us smarter on it and how it might impact you.

[blog] URL context tool for Gemini API now generally available. This is such a powerful feature of the Gemini API. Augment the static knowledge of an LLM with ingesting content from a URL. I like that it now supports PDFs, images, and data files.

[article] Your Company Needs to Focus on Fewer Projects. Here’s How. Any of you feel like you’re doing too few projects right now? Anyone? Bueller? No, most of us know there’s too many competing priorities. This is a good read on how to narrow your focus.

[article] Why software developers burn out, and how to fix it. Constant interruptions are killing your folks. That, and vague requirements are combining to burn out some great folks.

[article] The Top AI Tool for Devs Isn’t GitHub Copilot, New Report Finds. Surveys are surveys, but there’s signal here. It’s amazing to see how fast developer preferences evolve.

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

Happy Friday. Any weekend projects planned? My kids go back to school next week, so we’ve got one more good weekend of shenanigans before they have to go LEARN STUFF again.

[blog] Creating an Automated UI Test of Your Web App in Seconds, with Gemini CLI and BrowserMCP. There’s a LOT of potential here. It’s never been easier to do automated UI testing of your app.

[blog] Production-Ready Gen AI: Beyond the Python Prototype. Shots fired! Rod says you shouldn’t be focused on Python if you’re doing enterprise AI apps. The JVM is his jam.

[blog] How Google does it: Your guide to platform engineering. Some useful framing in this post from folks who see how we manage platforms ourselves.

[blog] This Week in Gemini CLI (vol. 2) 🔥. Great updates this week. I like the ability to mess with MCP servers from the CLI and the direct VS Code integration helped me out a couple of times this week.

[article] That ‘cheap’ open-source AI model is actually burning through your compute budget. Huh. That’s news to me. I didn’t realize open weight models use a lot more tokens than the proprietary counterparts.

[blog] Monitor your databases on Compute Engine with Database Center. Low-key awesome. Even if you self-manage your databases on cloud VMs, they can get the advanced monitoring and AI assistance we offer to managed databases.

[article] Should CIOs care about new AI models? It’d be easy to ignore the steady stream of new models (and versions) that come out. Don’t settle into an old one for too long though!

[blog] Cloud CISO Perspectives: New Threat Horizons details evolving risks — and defenses. Find some key takeaways from our biannual threat report. Credential compromise and misconfigurations are still a top entry point for threat actors.

[article] Building Resilient Platforms: Insights from 20+ Years in Mission-Critical Infrastructure. Applicable advice here from someone who has seen what works, and what doesn’t.

[blog] Start and scale your apps faster with improved container image streaming in GKE. This is even more important now as we load large models and don’t want slow startups.

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

Some thought-provoking opinions in a few of the items in the list below. I’m realizing how quickly my own opinions evolve as a saturate myself in different ideas.

[blog] Why Writing Matters (to the Technical Professional). From what I can tell, good communicators succeed more often than bad ones. Keep working on the craft of writing!

[blog] One Toolbox, Many Sources: Unifying Your Data Landscape with MCP Toolbox. We’re going to end up with MCP sprawl, yes? Seems inevitable. Solutions like this which let you talk to multiple databases via a single MCP server might be the right pattern.

[article] Top findings from the 2025 Stack Overflow Developer Survey. Good look at a few dimensions of how engineers work.

[article] What New CEOs Should Ask Themselves in Their First 100 Days. This is written for CEOs, but I thought it was useful for someone in any new leadership role.

[blog] The 5 Most Common REST API Design Mistakes (and How to Avoid Them). I’ve probably done all of these. This week alone. But good reminders that we should pay attention to.

[blog] Designing API Error Messages for AI Agents. I’ve never seen someone talk about this. What kinds of error messages should you return to help an agent figure out next steps?

[blog] Inferencing — Serve Llama 4 on A4 (B200 GPUs) using vLLM and GKE. Good instructions on serving an LLM on a fully-managed Kubernetes cluster.

[article] Can Platform Engineering Accelerate AI Adoption? The recent PlatformCon had a lot of discussion about platform engineering, and Jennifer pulls out many key insights.

[blog] Introducing Gemma 3 270M: The compact model for hyper-efficient AI. Here’s a flavor of our open model family that’s great for fine-tuning.

[paper] TextQuests: How Good are LLMs at Text-Based Video Games? Can modern models perform well when playing text-based video games? This research puts that idea to the test.

[article] GPT-5: A Choose Your Own Adventure for Frontend Developers. Fascinating launch of this model. Different experiences are still coming out. Theo revised his early impressions.

[blog] Designing a multi-tenant GKE platform for Yahoo Mail’s migration journey. There are some useful details in this story about Yahoo’s migration to the cloud.

[article] New trend: extreme hours at AI startups. There are usually seasons where you need to dig in and work extra. We’ve all been there. But making it your culture? Red flag.

[blog] How Keeta processes 11 million financial transactions per second with Spanner. Wow, some impressive numbers for a blockchain and payments business.

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

After yesterday’s MCP debacle, I built a new server using FastMCP and Python. All was right with the world. It was fun to create an API, expose it through tools in an MCP server, and then have my agent (the Gemini CLI) use those tools to invoke the API. I found it useful to build it all out!

[blog] Scaling down to speed up: How we improved efficiency of live process metrics by 100x. Datadog shares what they did, and it may inspire your own efforts to improve timeliness of data.

[blog] Gemini CLI Tutorial Series — Part 9: Understanding Context, Memory and Conversational Branching. Terrific post here on how to reset context, save and branch sessions, and other tips for our agentic CLI.

[blog] Gemini CLI + VS Code: Native diffing and context-aware workflows. This is big. I installed it earlier today, and it’s so handy to have a souped-up terminal when I’m working in the IDE.

[blog] AI Prototyping: How 11 Real-World Teams Are Transforming Their Work with Lovable. What are product people doing with AI prototyping tools like Lovable? I like seeing these real stories.

[blog] Secure your LLM apps with Google Cloud Model Armor. You’ll find a few ways to protect your LLM apps. This is an underrated service that can detect prompt injection attempts, block malicious URLs, and prevent sensitive data leakage.

[article] Unpacking METR’s findings: Does AI slow developers down? Some fresh perspectives from someone who participated in this buzzy study.

[blog] How to Stop Unauthorized API Access. Trying to keep people out of your APIs? This is a pretty good list of ways to do so.

[blog] Still Packaging AI Models in Containers? Do This Instead on Cloud Run. Powerful use case from Karl. You can stream the model files and keep it separate from your app code.

[blog] Go 1.25 is released. Nothing dramatic in this release, but some useful upgrades to tools, compilation, standard libraries, and garbage collection. Release notes here.

[blog] Gemini CLI or Jules: Which tool is for you? Great explainer. Yes, many tools have overlapping scenarios, even when coming from the same creator. But there are distinctions that matter.

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

Sometimes I choose the “hard way” for no good reason. I’m building an MCP server for educational purposes, and instead of choosing a full-featured library, I choose a capable, but loosely documented one. So I spend most of my time fiddling with ambiguous problems that even agentic AI isn’t helping me with. Forget sunk-cost fallacy. Tomorrow I start over with Python or TypeScript!

[blog] Why Can’t I Just Use an API? Because Your AI Agent Needs MCP. Strong post that explains why handing a pile of API endpoints to an AI agent isn’t the smartest strategy.

[blog] Supercharge your AI: GKE inference reference architecture, your blueprint for production-ready inference. This links to an executable blueprint with a performant, scalable, cost-effective, and observable Kubernetes foundation for AI inference.

[blog] You need evals to ship AI features. Enough folks are saying this that you should pay attention. If you’re going to be successful with AI, you’ll need strength at evals.

[blog] How Yahoo Calendar broke free from hardware queues and DBA bottlenecks. Detailed story about a careful move of critical infrastructure for a global service.

[blog] How I coded the Google Style Guide into a Gemini CLI Custom Commands Workflow. Our tech writing team is getting into the action. Shweta has a good post about using AI as a content checker.

[blog] Early Adopters Share AI-Centric Service Desk Results. There’s no debate that AI can make summarization easier. And in some cases, help close very routine support cases.

[article] Multi-agent AI workflows: The next evolution of AI coding. Take it one at a time. Don’t thrust a dozen agent personas into your team and expect anything except chaos.

[blog] The Google Developer Program is evolving. The perks for the freebie edition are good. If you want more things (and a good value), there’s now a paid Premium tier where you get more AI credits and higher usage limits, along with cert vouchers.

[blog] Bottleneck or Bisect: AI-Assisted Coding Will Change Product Management. Good post. A subset of PMs are going to need to adapt to faster engineering turnarounds and public experiments.

[blog] Claude Sonnet 4 now supports 1M tokens of context. I’ve been somewhat shocked that it’s taken this long to get another model to 1 million input tokens. Google Gemini’s had this for quite a while, but it requires some sophisticated models and underlying infra to support.

[blog] Google is a Leader in the 2025 Gartner® Magic Quadrant™ for Container Management. Your choice of container management runtime(s) matters a lot. This year’s assessment was super thorough, and you should look through the report.

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

Did your week start off fast, or are you able to ease in? I had a mix of both. Today’s reading list has the requisite amount of AI content, but also some non-AI content that I think you’ll like.

[article] Do not Interrupt Developers, Study Says. Keep devs in a flow state. Reduce meetings and interruptions if you’re actually serious about productivity!

[blog] Coding for the Future Agentic World. Take a couple of minutes and go through this. It’s worth it. Addy wrote an excellent tour of the tooling and practices that make up today’s agentic coding world.

[blog] Goodbye API Keys: Gemini CLI GitHub Actions with Workload Identity Federation. API keys are easy to use, but not a great long-term strategy. This is a good post about creating a more responsible security story.

[blog] The Future Isn’t Model Agnostic. I guess this qualifies as a hot take, but it shouldn’t. Model swapping isn’t trivial and you end up optimizing for the wrong things.

[blog] Seamless Istio Upgrades at Scale. Good story. Anyone can deploy things nowadays. It takes planning and skill to cleanly upgrade what’s been deployed.

[blog] Using GKE to Build a Developer-First Internal Platform. Kubernetes is boring now. Which is great! That means it’s widely adopted, mature, and trusted for a variety of workloads.

[article] The Future of Product Management Is AI-Native. I liked the insights here, especially around shared AI work among team members.

[article] Kubernetes Will Solve YAML Headaches with KYAML. Were you aware of this? I was not. There’s a new subset of YAML coming to Kubernetes.

[blog] Run OpenAI’s new gpt-oss model at scale with Google Kubernetes Engine. And go ahead and run OpenAI’s new open model on our Kubernetes. That’s not a sentence I thought I’d write two months ago.

[blog] The Most Underrated Performance Enhancer: Having Fun. I tend to say that I’m not a serious person, but I’m very serious about my work. Tense leaders make tense teams.

[article] The advantages of stack-based internal developer platforms. This is an approach where devs can self-service infra from a set of templates that include all the components for shipping and managing apps.

[blog] Why OpenAPI Should Be The Foundation For Your MCP Server. Use an OpenAI spec to bootstrap your MCP server? I can see the merits, but it’d be easy to accidentally do a 1:1 mapping of operations to tools, without thinking through higher order tool types.

[article] GitHub CEO To Step Down As Company Is More Tightly Embraced by Microsoft’s CoreAI Team. I suspect this will lead to a ripple effect. Any attempt to retain independence within GitHub seems like it’ll be a challenge. More here.

[blog] Gemini CLI Tutorial Series — Part 8: Building your own MCP Server. Speaking of MCP servers, this is a very good post about creating, testing, and using your own.

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

It was a busy week, and I’m happy for the weekend to be here. Time to recharge. That doesn’t mean no tech work; just more of it on my own terms!

[blog] New Gemini app tools to help students learn, understand and study even better. My first kid goes to college in a couple of weeks, and these sorts of tools will likely become standard in the years ahead.

[article] Your AI Strategy Needs More Than a Single Leader. Don’t fall into the lone-leader trap. Distribute leadership and ensure many executives, including the CEO, act as stakeholders.

[article] High costs and thin margins threatening AI coding startups. Free tools are expensive to somebody. Right now, its the vendors and their VC backers.

[blog] Prepping Your Agent for Production — Observability for Agent Development Kit (ADK). Good content here for those looking to build robust agents that are maintainable.

[youtube-video] Shift down: A practical guide to platform engineering – Leah Rivers & James Brookbank. A couple of my smart colleagues share useful thoughts based on how Google works, and what our customers do.

[blog] Gen AI Evaluation Service — Agent Metrics. I’ve said this a few times here, but I really do believe you want to gain familiarity and skill with AI evals. Mete looks at agent evals. Also, a follow-up post on multimodal evals.

[article] How does the SPACE framework apply to AI’s impact on engineering? It’s not just about measuring velocity gains with AI. What’s the overall impact on engineering?

[article] Google Launched LangExtract, a Python Library for Structured Data Extraction from Unstructured Text. It’s open source, and looks like a useful way to get info out of large bodies of text.

[article] Businesses unprepared for AI agents: EY. Sure. Rarely is any company completely prepared for disruptive technology. But the smart companies are quickly addressing the fundamentals.

[blog] Google is a Leader in the Gartner® Magic Quadrant for Strategic Cloud Platform Services. Hey, look at that. The most visionary cloud, with execution greater than our friends in Redmond? Quite the progress report for us.

[article] DevOps Pioneer: Vibe Coding 100x Bigger Than DevOps Revolution. What “vibe coding” means seems to depend on the person. This upcoming book uses it in the title (I read an early draft), but promotes a more responsible style of vibe coding.

[article] ChatGPT users hate GPT-5’s “overworked secretary” energy, miss their GPT-4o buddy. Yes, change is hard. And whether a model is “better” or not may be relative for some, nowadays. Related. And more. Sounds like OpenAI will rollback the removal of 4o.

[blog] Build Better Tools, Faster: A Deep Dive into the New MCP Toolbox UI. I can’t only work in a terminal. This visual UI helps devs experiment with their MCP interactions.

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

Big-ish reading list today, and my queue is still stubbornly bloated. There’s a lot going on in our industry, so I don’t expect a slowdown in content any time soon. I hope these roundups save you some stress in trying to keep up!

[blog] Your AI workloads still need a service mesh. “Need” might be a strong word, but “can make use of”, definitely. Good post by John.

[article] Why the AI era is forcing a redesign of the entire compute backbone. Hardware, networking, power, and more are subject to evolution when you deeply invest in AI.

[blog] Secure your storage: Best practices to prevent dangling bucket takeovers. Yeesh, this refers to an attack vector I hadn’t thought of. Is your code referring to cloud storage buckets that were deleted? What happens if someone else takes that name/URI?

[blog] How Tech Companies Undermine Their Content Marketing. Smart guidance, which could apply to content creation for any (internal/external) audience.

[blog] Introducing GPT-5. Congrats to OpenAI on a big launch day, with well-coordinated support across a lot of popular surfaces. Well done. Ethan and Simon both take a deeper look and conclude that it’s good, competent, but not transformational.

[blog] Redefining enterprise data with agents and AI-native foundations. You can do a ton with a great model. But I get excited about the experiences on top that help me do interesting things. Here’s a great case of value atop AI models.

[blog] The latest AI news we announced in July. I’m liking the breadth of useful capabilities we’re getting into the hands of users.

[blog] Spanner columnar engine: Powering next-generation analytics on operational data. Maybe one database engine is all you need? Cloud Spanner now makes it simpler to do analytical queries against operational data.

[blog] How Google and NASA are testing AI for medical care in space. If history repeats itself, innovations discovered here will make their way to everyday experiences for humanity.

[blog] Go at American Express Today: Seven Key Learnings. Go isn’t just for tool builders and leading edge companies. It’s for the most demanding environments too.

[blog] Interesting Links and Observations, July 31st, 2025. My friend Fintan shares link roundups from time to time, and this one has references to half-a-dozen dev surveys that are worth reviewing.

[article] Getting started with A2A in .NET. Cool to see Microsoft contribute an early preview of a .NET SDK for A2A. Building multi-agent solutions is getting simpler.

[blog] Vibe-coding a Chrome extension with Gemini CLI to summarize articles. I’m fairy sure that Guillaume wouldn’t describe himself as an “extension developer” but nowadays, anyone can build anything. Great story here.

[blog] Atlassian partners with Google Cloud to power a multi-cloud future. Glad to see Atlassian hosting some of their most important properties on Google Cloud.

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

Good day back at home after a day trip. You’ll find at least four pieces below that have some strong opinions about software development. That’s what I like reading!

[blog] No, AI is not Making Engineers 10x as Productive. Good post. AI helps, it’s not magic. There are areas it makes a big difference, but it’s not the entire workflow. Yet. Or maybe never.

[blog] Meet your new AI coding teammate: Gemini CLI GitHub Actions. There’s value in adding AI to *more* of your workflow. Here’s a free addition to your GitHub repo that can triage issues, perform code reviews, and even implement proposed changes. More here.

[blog] This Week in Gemini CLI. That GitHub feature is just one of a few very nice updates to the CLI. I really like the ability for a CLI session to reference multiple file directories.

[article] OpenAI launches two ‘open’ AI reasoning models. Big model week so far, and it’s only Wednesday. Good to see some open work from OpenAI.

[article] Anthropic’s new Claude 4.1 dominates coding tests days before GPT-5 arrives. Claude continues to be a great choice for coders. They’ve got the mindshare with developers, and have earned it.

[blog] Vibe Coding vs. Spec-Driven Development. I really like the spec-driven approach as it steers us back towards decision records and thoughtful planning before setting agents (or humans) loose.

[article] Amazon will offer OpenAI’s open-weight models, sidestepping Microsoft via Apache 2.0 license. Smart of OpenAI to continue diversifying its availability. AWS needed this too!

[article] Wells Fargo taps Google Cloud for agentic AI push. Value is moving up the stack. Models are necessary but not sufficient. Agentspace (now used by Wells Fargo) makes AI actionable for everyone. More here.

[article] Refactoring Stubborn, Legacy Codebases. Vibe coding new stuff is cool, but we all know that the gnarly work happens with existing codebases. Stripe engineers talk about how they worked through a modernization.

[blog] Jules, our asynchronous coding agent, is now available for everyone. This background agent for GitHub is now GA. Simon likes the term “asynchronous coding agent.”

[article] Stop Hunting 10x Engineers: Build a 10x Organization Instead. Discouraged because you can’t recruit 10x engineering unicorns? Stop it. You can likely build a fantastic org with the talent you have available.

[blog] Optimize your cloud costs using Cloud Hub Optimization and Cost Explorer. App ownership isn’t always glorious. Especially if you’re digging around logs and billing spreadsheets to figure out chargebacks. This looks like a step up.

[article] JetBrains previews no-code app builder. Good complementary offering from an established IDE vendor.

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