Author: Richard Seroter

Heading into what I hope is a lowkey weekend. I’m excited to not do a whole lot. May your weekend be as action-packed or boring as you’re hoping for.

[blog] In Praise of RSS and Controlled Feeds of Information. At least 80% of my reading list comes from the RSS feeds I follow. The rest is from more real-time sources, newsletters, or random links I come across.

[blog] A collaborative approach to image generation. Not quite getting the image you’re after? The Google Research team looks at a new reinforcement learning agent that helps refine images by learning user preferences.

[article] What’s AI’s impact on software delivery performance? (2025 DORA Report). Analysis of our recent report is trickling out, and I thought this piece bubbled up some key findings.

[blog] This Week in Gemini CLI (vol. 5). Very good updates this week. I really like the potential of a managed todo list of agentic tasks.

[article] When Managing Your Team Becomes Too Much. There are times that I feel this. I manage a big (awesome) group and it can be overwhelming. Good advice here.

[blog] AI Won’t Replace Design Engineers, But It Will Change How We Work. Actively define how (or if, or where) AI complements your work before someone else does.

[blog] The History of Core Web Vitals. I knew basically none of this. I remember AMP and such, but didn’t realize all this other work happened. Impressive!

[article] Top CIO conferences to attend in 2026. Not a bad list of events to consider if you’re looking to make smarter strategic choices.

[article] Who stops wasteful cloud spending? “Waste” is inevitable, right? It’s difficult or impossible to be fully utilized. But aim for better! That said, how much on-prem spending is wasted? I’d argue it’s more 🙂

[article] Generative AI drives cloud spend blitz. You’re going to spend more in the cloud, because it offers the best way to take advantage of AI tech.

[article] The EM’s guide to AI adoption (without your engineers hating it). Here’s a good playbook for incorporating AI into your engineering team in a way that sticks.

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

I’m grateful that the most persistent stress in my life comes from following my favorite sports team. There are worse things in life. But one of these years, I really would like my hometown team to, just one time, raise the trophy.

[blog] Gemini 2.5 Flash Image now ready for production with new aspect ratios. The best image model in the world is now GA. These new aspect ratios are terrific.

[blog] Building on the bananas momentum of generative media models on Google Cloud. What stood out to me here was a bulleted list at the bottom that clearly explained when to use which model. We’re getting better at doing that.

[blog] Inside Husky’s query engine: Real-time access to 100 trillion events. That’s a lot of events. The Datadog team explains their event store and how its query engine works.

[blog] Gemini CLI: Discover, configure, debug, and document a Google Cloud feature (Part 2). These AI tools can help everyone learn new things. One of our technical writers, Moi, explains how she used the Gemini CLI to go deeper on a product area she’s writing about.

[blog] Two strategies to succeed when AI seems to be eroding jobs around you. This is relevant to the above post. Tom’s a tech writer here at Google, and has great perspective on what skills need further investment.

[blog] Real AI Agents and Real Work. We’re learning more about human-agent interactions. Agents are capable, and still require our judgement to decide what’s worth doing with them.

[blog] Examples are the best documentation. Great post. People who deliver projects and products need to consider all the examples the user needs access to.

[article] OutSystems Launches a Low-Code Workbench for Building Enterprise AI Agents. Of course low-code vendors will get into the vibe and agent game. They’re irrelevant if they don’t!

[article] Salesforce launches enterprise vibe coding product, Agentforce Vibes. See previous point. This seems natural for Salesforce, as they’ve got piles of data for you to ground on.

[blog] Introducing Microsoft Agent Framework: The Open-Source Engine for Agentic AI Apps. Do we need more agent frameworks? Sure, why not. This one targets .NET devs.

[blog] Designing agentic loops. What does a responsible YOLO mode look like when using your coding agent of choice? Simon’s got a helpful post on designing good agentic loops.

[blog] Meet Jules Tools: A Command Line Companion for Google’s Async Coding Agent. This is cool. Control and interact with this remote agent via the CLI.

[blog] Angular support for generating apps in Google AI Studio is now available. Excellent! It’s now simple to vibe code web apps that use Angular and then deploy them.

[blog] Build Your Foundation First: The Hard Truth About Successful AI Deployments. No skipping ahead. There are prerequisites required to get good AI results.

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

Had a day-trip to Sunnyvale to present to a room full of Google Cloud partners. Wasn’t feeling 100%, but still seemed to go ok. I referenced something from today’s reading list, which proves to me that I should keep doing the work to uncover these pieces 🙂

[blog] Effective context engineering for AI agents. Exceptionally good writeup that helps us better understand what context engineering is all about, why it matters, and how to approach it.

[blog] Cloud CISO Perspectives: Boards should be ‘bilingual’ in AI, security to gain advantage. I think a lot of non-engineering teams are thinking about how to become more fluent in AI so that they can make better decisions.

[article] Boards lack AI savvy as adoption advances: PwC. I can see why the above advice is so important. Most board members are flying blind.

[paper] On the Use of Agentic Coding: An Empirical Study of Pull Requests on GitHub. I haven’t seen a study like this one. It explores pull requests created by Claude Code, and how agent-generated PRs differ in size, acceptance rate, and such.

[blog] 100X Faster: How We Supercharged Netflix Maestro’s Workflow Engine. There’s probably no action item for you after reading this post, but it’s ok to just enjoy learning about what other people built.

[blog] Responsible Agents: A Phased Approach on Google Cloud. Excellent post. Here’s a well-rounded look at making responsible decisions during design, development, evaluation, deployment, and operations of agent systems.

[blog] Securing MCP Servers with Spring AI. Learn how to build MCP servers, but pay special attention to how you secure them too.

[blog] Forecasts and data insights come to BigQuery’s MCP and ADK tools. Or use turnkey MCP servers that already have great tools available in a secure fashion.

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

Spent some free time today building an AI agent with persistent memory. Everything’s not working yet, and it reminds me how annoying out-of-date LLMs can be. Hard to get accurate advice on new technologies!

[blog] The common sense unit of work. Many things—how you plan, what you report, your approach to AI—depends on what you consider a unit of work.

[blog] The Agency Spectrum: An AI Risk Management Framework. This makes sense to me. What levers of governance do you have at your disposal when building an agent system?

[blog] Your employees are already AI-augmented. Your enterprise isn’t. Here’s why that matters. I’ve seen a few pieces on this, and observed this in real life. People are using whatever tools help them. That doesn’t always align with corporate initiatives or buying patterns.

[article] Legacy tech woes fuel IT upgrades as CIOs grapple with costs. Somewhat related to the previous entry. Most enterprise leaders are navigating dense portfolios, while employees figure out how to get today’s work done.

[article] Flow State to Free Fall: An AI Coding Cautionary Tale. Great analogy, and many of us who work with these tools understand the pain of climbing too fast with AI.

[blog] Everyone Wants To Grow — Few Want To Do The Work. You won’t advance—or see your team advance—without intentional investment. Many folks talk a good game, but not many put in the work.

[blog] What is “good taste” in software engineering? Great post. It takes time to develop “taste” and by the time you have the experience, you might be too stubborn about “best practices” to adapt!

[article] How do developers collaborate with AI agents when solving real-world software engineering problems? It’s probably confirmation bias, but this data affirms what I’ve been seeing and believing.

[blog] Building a Next-Generation Key-Value Store at Airbnb. You should buy versus build in about 96% of scenarios here, but it’s still cool to see what Airbnb built.

[article] The YouTube Tip of the Google Spear. Fascinating writeup of what’s still an under-valued service: YouTube.

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

I’m mostly adjusted back to Pacific Time. I had a great weekend with visiting family and was ready to get going today.

[blog] Creative Java AI agents with ADK and Nano Banana. Cool post. How do you build an agent that can process image results and then use those in further agentic steps?

[blog] Give your AI eyes: Introducing Chrome DevTools MCP. This will be a top-3 MCP server for a lot of web devs. Check out this post for all the ways you can bring Chrome actions into your agent or IDE.

[blog] Introducing Claude Sonnet 4.5. Impressive update from Anthropic, and some big statements. They’ve also updated their coding tools. You can use their latest model on all the clouds, including Google Cloud and Vertex AI.

[blog] The AI coding trap. I’m learning a lot from people who are sharing their experience using AI within real engineering teams.

[blog] Development Productivity, Not Developer Productivity. Measure and improve the whole system (or at least, the actual bottlenecks) if you want to see tangible improvements.

[youtube-video] What is GenKit Go? This framework (for TypeScript, Go, and Python) is underrated, but also fairly popular among those who want a clean way to build AI apps.

[blog] What are popular AI coding benchmarks actually measuring? Educational post for those who want to understand what’s good (and bad) about the current crop of coding benchmarks.

[blog] Flight Recorder in Go 1.25. I’m very impressed by the thoughtfulness of this built-in solution for grabbing trace data when something goes wrong in my Go app.

[blog] 90%. Here’s another terrific experience report. AI tools are writing 90% of Armin’s code on recent projects, but that’s because he’s providing a strong guiding hand.

[blog] Documentation. Another short title, but another strong punch. Lee offers up then traits of incredible developer documentation, and it reflects many of the things my team is chasing now.

[article] Why the Frontend Should Embrace Platform Engineering. I haven’t seen this perspective before, and it’s good to see a focus on serving all types of devs with our platforms.

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

And … I’m home. Looong day of travel from South Korea to LAX before a 23 minute flight to San Diego. Uneventful, wifi-free, so I was forced to read and relax. Enjoy your weekend.

[article] Getting AI to Work in Complex Codebases. Wow, this might be the most interesting piece I read all week. Check this out for a tried-and-tested workflow for using AI in big complex codebases.

[article] Who leads in AI adoption: Tech or non-tech? Are enterprises going faster in their adoption of AI tools than those in BigTech companies? Maybe. But data also shows BigTech ships faster, which is what actually matters.

[blog] Generating Consistent Imagery with Gemini. Neat idea! Can you breathe new life into exist image libraries? Laurent offers a very visual example.

[article] InfoQ AI, ML and Data Engineering Trends Report – 2025. What is mainstream, and what’s emerging in the AI/ML and data engineering space? The InfoQ team takes a look.

[blog] Gemini Robotics 1.5 brings AI agents into the physical world. Are we going to see an explosion in robotic advances that mirror what’s happened with AI? Maybe so!

[article] To vibe or not to vibe. It depends. I half wonder if this practice will melt into standard workflows and be something we employ in a task-by-task basis.

[article] Oracle saddles up with $18B debt amid AI infrastructure gamble. Oracle has been racking up some impressive wins with massive contracts. They don’t seem to have the infrastructure to satisfy their commitments, but sounds like they’re working to resolve that.

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

Today’s my birthday, and I celebrated with a quiet final day in South Korea. It’s been a good week, but I’m definitely eager to get back home. Should have a reading list for you tomorrow as well!

[blog] Meet the new GKE: Extending Autopilot to all qualifying clusters. Terrific. The best way to run Kubernetes in the cloud (Autopilot mode in GKE) is now available to any GKE cluster.

[blog] Deploy your own AI vibe coding platform — in one click! These folks do some fascinating things. Cool company. Here’s an open source vibe coding platform that uses Gemini by default.

[article] How Google’s dev tools manager makes AI coding work. My friend and coworker Ryan is featured here talking about AI coding tools and the future of development.

[blog] CRUD-Sourcing is why Your Event Streams Are Bloated. I like this point. And it applies to MCP servers, and all sorts of things. Stop modeling everything as CRUD operations. Think about event stream sessions and tools in context.

[blog] Introducing the Data Commons Model Context Protocol (MCP) Server: Streamlining Public Data Access for AI Developers. Provide trusted data to your LLM as context by using this new MCP server for the Data Commons dataset. It’s a long-running aggregation of the world’s public data. Press coverage here, and here.

[article] Is Seattle’s tech scene in trouble? WSJ report highlights concerning trends — with a potential opening for startups. When I lived up in the Seattle area, I thought the region was invincible to downturns given the critical services based there. I was wrong, and there’s definitely a change going on!

[article] It isn’t your imagination: Google Cloud is flooding the zone. A lot of strategery going on right now among the big players. I think we’re making some shrewd moves.

[blog] Launching Gemini CLI extensions for Google Data Cloud. Create a more structured an predictable way to bring cloud data into your CLI sessions. Create databases, perform analysis, and more.

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

Today was the big day for the South Korea trip. I was with our friends at LG who are hosting their 3-day developer conference. After doing a keynote, I had the pleasure of meeting with various executives, and answering questions about how to increase productivity with AI. Good timing, with the new DORA report out today!

[blog] Announcing the 2025 DORA Report: State of AI-Assisted Software Development. An absolute must-read this year. I’m going through it now, and it’s 140+ pages of tremendous data and analysis.

[article] Six Frameworks for Efficient LLM Inferencing. Once you have a model, you need an inferencing engine to serve predictions with it. Jani offers up a helpful review of your options.

[blog] ​​Don’t Fire Your Developers! What AI-Enhanced Software Development Means For Technology Executives​. What he said. But looks like few managers are thinking that way.

[article] Survey Surfaces Software Engineering Challenges Following Adoption of AI. Our DORA research shows this as well. You don’t just drop AI tools into a team and magically get better performance.

[article] AI-Generated “Workslop” Is Destroying Productivity. Here’s the flip side of all this AI enthusiasm. Will we churn out more mediocre content in the name of “productivity” and skipping real thought by outsourcing tasks to AI? Guard against that.

[blog] Strengthening our Frontier Safety Framework. I’m glad that we constantly re-evaluate this and update it with what we’ve learned.

[blog] The Myths (and Costs) of Running Node.js on Kubernetes. Interesting take. The pitch here is for a more intentional approach to using k8s to run Node, assuming you even need to run it there.

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

Tuesday done here in Seoul, and I enjoyed talked to a couple of customers today. Big focus on how to get quality results from LLMs. No single solution to that problem, but a mix of model, smart context, and grounding can help.

[blog] Memory Allocation in Go. Very few people need to know all of this. Probably like twelve. But I admire the depth of content, and learned a few things.

[blog] How to Choose Your AI Agent Framework. Great post. The author looks at different types of frameworks, and offers up a decision tree for choosing one.

[report] Anthropic Economic Index report: Uneven geographic and enterprise AI adoption. Read this online report from Anthropic to get a sense of where there are concentrations of AI usage, and what those leading populations are doing.

[blog] If you are good at code review, you will be good at using AI agents. What an interesting point. Sean makes a reasonable argument here, and his experience mirrors what I’ve seen (regarding the agents) as well.

[blog] How we built it: Real-time analytics for Stripe Billing. Ideally, your systems would tell you immediately what the cost was for the operation that just happened. Easier said then done. But Stripe rebuilt their billing pipeline and got latency down to 15 minutes. Impressive!

[blog] How 18 companies measure AI’s impact in engineering. Read the newsletter this came from, but this post gives an overview and a visual that you’ll find useful.

[blog] Enterprise AI Looks Bleak, But Employee AI Looks Bright. People use LLMs where it helps them get something done. It’s interesting to see the disconnect (again) between executives and employees.

[article] Identify Your Core Values to Make Better Leadership Decisions. Even if you don’t think you have any “core values” that drive your behavior, you do. There’s SOME heuristic you’re applying to decision making.

[blog] We’re Optimizing Ourselves Into Oblivion. The summary Cote provides here spoke to me. Everything doesn’t have to be efficient.

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

First workday in Seoul is complete. I tried to wear myself out on Sunday by walking around for hours, but still somehow woke up wide-awake at 3am. But a great day, and customer visits up tomorrow. Meanwhile, a fun reading list!

[article] The Purpose of Prototypes. What’s the purpose of a prototype? Your answer may be different from Marty’s, and it’s worth reviewing his.

[article] Is Your Startup Idea Any Good? Borrow These Validation Tactics from the Founders of Linear, Mercury and More. Somewhat related to the previous post. How are you vetting your ideas and finding out whether it’s worth building out?

[blog] The quality of AI-assisted software depends on unit of work management. Read this. I think it presents such a strong point when thinking about AI’s role in software dev, and how to find the right unit for work.

[blog] How Agents Think. Do LLMs “think” like a human brain does? I don’t think so. But patterns like the ReAct loop combined with reflection looks a LOT like how we approach problems.

[blog] AI Was Supposed to Help Juniors Shine. Why Does It Mostly Make Seniors Stronger? I get the hypothesis. Our own data isn’t bearing this out. Earlier-career devs seem to be using it more often to good results.

[blog] 2-Minute ADK: Speed Up Your Agent with Parallel Tools. Simple post, but great reminder to optimize everywhere.

[blog] I think “agent” may finally have a widely enough agreed upon definition to be useful jargon now. Saying “an LLM agent runs tools in a loop to achieve a goal” is a solid way to describe what an agent is.

[blog] This Week in Gemini CLI (vol. 4). For those who don’t want to wade through release notes, posts like this are super useful.

[blog] Achieve agentic productivity with Vertex AI Agent Builder. Everyone is going to be offering you platforms to build and run agents. Be clear on what you’re looking for.

[blog] How to Evaluate AI Coding Tools for Your Enterprise. It’s tough to mentally organize all the different AI tools out there, so this categorization works as well as any.

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