Author: Richard Seroter

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:

Had a quick day-trip up to Mountain View and back. Meetings were had. And content was still read.

[blog] Genie 3: A new frontier for world models. Generate dynamic, explorable worlds with nothing but a text prompt? And these worlds render in real time at 24 frames per second? Remarkable.

[article] Why Leaders Should Bring Their Best Self—Not Their Whole Self—to Work. I’m not a fan of the “bring your whole self to work” or “just be yourself.” I try to present my best self without being a weirdo.

[blog] How to Use BigQuery Sandbox with Colab for Data Science. Query loads of data at no cost? Or without sacrificing a credit card? Jeff shows you how.

[article] The Case Against Metrics for Developer Productivity. You want some way to measure team performance besides gut feel. But an heavy reliance on sterile dev metrics is risky.

[blog] Create personal illustrated storybooks in the Gemini app. I love this very much. Whether you have kids, teach kids, or just want a cool way to tell a story, this is such a cool feature.

[blog] Memory in Agents, Make LLMs remember. Agents without tools or memory are going to feel tired quickly. Philipp explains more about your memory options.

[article] 9 habits of the highly ineffective vibe coder. Some good anti-patterns here! We can learn a lot by observing what does NOT work well.

[blog] Introducing Google Cloud Setup: Your guided pathway to a secure cloud foundation. It shouldn’t take so much work to set up your cloud the right way. This effort has been years in the making, and taken enterprise-grade setup time from weeks to minutes.

[blog] Building A Story Index Of A Decade Of TV Evening News With Geocoding, Sentiment, Frame & Narrative Analysis In 1 Hour For $154. Wow. Processing fifteen years of broadcast news and making an index of all the stories? For $150? Amazing to see what’s possible.

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

Back at the office today after a vacation week. I admire my colleagues who “select all: delete” their inbox on the first day back. I don’t have that level of courage.

[article] When to Innovate and When to Imitate. In many cases, you should just copy what’s already out there. This article offers advice on when you should which strategy.

[article] AI use among software developers grows but trust remains an issue – Stack Overflow survey. I still need to dig through the survey itself, but we can read this piece to learn about what developers think of AI. More here.

[blog] Friday Forward – Different Strokes. Good stuff. Know your priorities, and own it. Try and enjoy the process versus obsessing over the goals themselves!

[article] Google Brings the A2A Protocol to More of Its Cloud. This protocol (and associated tooling) is maturing for everyone. And, we’re making it simpler to use with Google Cloud too.

[blog] Don’t hide behind AI to trim your belly fat. Start redesigning your workforce. These are some of the most frank analysts that I come across. No punches pulled.

[blog] Concerns beyond the accuracy of AI output. Here’s a post from our DORA research team that uses data from extended developer interviews to learn what they worry about.

[blog] Gemini CLI: Power up your Linux workflow. These agentic CLIs do more than mess with code. You might use it to set up your local machine, analyze files, and more.

[blog] Disposable Code Is Here to Stay, but Durable Code Is What Runs the World. I read whatever Charity writes, and this one is worth you reading as well.

[blog] Remember this: Agent state and memory with ADK. Terrific post that clearly explains the ideas of short term and long term memory with AI agents.

[blog] Security for Agents and Agents for Security: The Next Cybersecurity Frontier. This is a good space for entrepreneurs. When you have VCs begging for founders, you know there’s a market.

[article] The Observability of Observability. With AIOps, who is watching the watchers? This post encourages an approach where you have checks on your AI automated systems.

[article] Firebase Studio Has Me Second-Guessing My Opinion on Developing With AI. There’s no shortage of these prototyping IDEs. Firebase Studio is a good choice and does a surprising number of different things.

[blog] Rethinking how we measure AI intelligence. It’s not just who can remember what. Let’s test these LLMs using more sophisticated approaches. Like this!

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

Happy Friday. Vacation is over for me and I’m back to work on Monday. I had fun taking my kids around town, and enjoyed some extra free time. Make sure you take your vacation hours too!

[blog] Try Deep Think in the Gemini app. It’s a variation of our recent gold-winning model and does parallel thinking through a problem.

[blog] Test-driven development with AI. I like the advice here to think about goals, not just the tests. And, good tip about AI to help with visual tests.

[blog] Is JSON prompting effective? (No.) Many folks (me included) feel this placebo effect when getting a tip on how to prompt AI for better results. “Oh that worked better!” Probably not?

[blog] Google Earth AI: Our state-of-the-art geospatial AI models. The mashup of AI capabilities with valuable datasets results in some super experiences.

[blog] Executives should be the least busy people. Definitely truth to this. Or at least, execs should have the most flexible calendars. Being booked solid is a sign of bad prioritization.

[article] Box CEO Aaron Levie sees a bright agentic future in the enterprise. Aaron posts on X some of the most pragmatic and insightful takes about agents. Here’s some additional perspective.

[blog] A deep dive into code reviews with Gemini Code Assist in GitHub. This no-charge code review agent that sits in your GitHub workflow adds a lot of value. I’ve used it a few times and came away impressed.

[blog] 2025 Mid-Year LLM Market Update: Foundation Model Landscape + Economics. Who is using what, and how is open model adoption going? And why do enterprises switch models? New data here.

[article] C-suite leaders attribute revenue, software development boosts to AI. I’m not sure about the results, or if there’s a disconnect between exec perspectives and employee ones. But, sometimes perception is reality!

[blog] How we revamped our Docs for AI. If bots, not people, read your docs, what should they look like? Provocative question!

[article] Research: The Hidden Penalty of Using AI at Work. Interesting. Learn how much the “competency penalty” is for using AI at work and how to resolve it.

[article] Hard-won vibe coding insights: Mailchimp’s 40% speed gain came with governance price. I’ll keep seeking out any pieces from folks explaining their real experiences with AI rollouts.

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

Vacation (at home) with kids is fun, but I’ll need another vacation after this. Today we went to the Del Mar racetrack to watch the horse races. And I’m a big baseball fan, so I was tense watching today’s trade deadline. The San Diego Padres were silly active.

[blog] Being a good PM at Google. Marvelous piece from Chris that applies to anyone wanting to be a premier product manager, anywhere.

[blog] How to evaluate vibe coding tools for your enterprise. This post attempts to categorize the current crop of AI tools, who they help, and what enterprises should look for.

[blog] An Engineer’s Guide to AI Code Model Evals. I see a lot of people use feelings to describe their opinions about AI-assisted tools. Feels fast. Feels slow. Feels like it gets worse. Addy looks at what a more objective eval of an AI coding model looks like.

[blog] Evaluating Agents with ADK, Part 1: The development loop with the ADK Web UI. Let’s keep the evals conversation going. Ivan covers the manual steps of assessing the flow of an AI agent.

[blog] Don’t Tell Engineers What to Do. It’s not a free-for-all. You don’t get to do whatever you want. But managers should be setting priorities (what to work on) and empowering the team to solve things the most responsible way.

[blog] Announcing a complete developer toolkit for scaling A2A agents on Google Cloud. Lots of updates for those who want to build and distribute agents.

[blog] How to Build an MCP Server with Gemini CLI and Go. Some fantastic lessons here from Daniela as she leaps hurdles and documents how she made the jumps.

[blog] How to Secure MCP Servers. If you must use MCP servers, then look at advice such as this to set them up securely.

[blog] AI is Eating Developer Experience. It’s being added everywhere to it, yes. If used to make things better (versus using it to use it), the team will thank you.

[blog] Gemini CLI: Custom slash commands. Here’s a simple way to access reusable prompts from our agentic CLI tool.

[blog] Pivoting From Software Engineer in Google, My First 3 Months. I like Annie’s story and have enjoyed her work so far. Here, she looks at what’s different when moving from software engineering into a developer relations engineering role.

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

It was another outside day of vacation. Kids and I toured the USS Midway aircraft carrier which is pretty terrific. It may be summer vacation time in this hemisphere, but there’s seemingly no slowdown in this industry.

[blog] What’s new in NotebookLM: Video Overviews and an upgraded Studio. We’re in a golden age of learning right now. It’s never been easier to study and digest new ideas. Love the video overviews!

[article] How to excel in multicloud: The new checklist. It’s a good list of things to consider. That said, there is no “easy” way to be multicloud, and most such solutions rely on a lowest common denominator approach, which strips all the value from your cloud strategy.

[blog] The Commodity Cloud Era is Over – The AI-Native Cloud Is Here. This is one reason you have to rethink your multicloud approach. There’s a good chance you’re solving yesterday’s problems and will be disappointed when you deliver solutions that aren’t relevant anymore.

[blog] Mastering agentic workflows with ADK: the recap. Great summary of some agentic workflow options and when to use which.

[article] Team got cut. Scope didn’t. Common story nowadays? What are you going to do about it? Here’s some advice that I’ve tried to follow myself.

[blog] Gen AI Evaluation Service — Tool-Use Metrics. I’m gaining new appreciation for this eval dimension. As you hand agents more tools, you want to be able to evaluate whether they’re being used correctly.

[article] METR’s study on how AI affects developer productivity. Abi takes a look at that surprising study from a few weeks ago showing that AI tools were slowing people down, not speeding them up. More study needed!

[blog] Pattern Matching with SQL: Real-Time Anomaly Detection Using BigQuery’s MATCH_RECOGNIZE. I didn’t know about this! Something that feels like a regular expression across rows in the data warehouse? Neat.

[blog] Kingdoms and Courts: The “Windsurf Effect” and How We Get Back to Normal in Tech M&A. Good post. I liked the articulation of three areas where AI-focused startups (and others) have moats.

[article] Google Cloud Q2 Surge in Revenue and Backlog Is Microsoft’s Worst Nightmare. Microsoft had terrific quarterly results. But our fundamentals might have a few competitors feeling nervous.

[blog] BigQuery meets ADK & MCP: Accelerate agent development with BigQuery’s new first-party toolset. You can build some valuable interactions with your data platform when its exposed as tools to an agent. This post has useful details on how to do that.

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

Another vacation day with the kids, this time to Sea World. Before everyone woke up, I got some productive reading time.

[blog] Veo 3 and Veo 3 Fast are now generally available on Vertex AI. This model has some amazing features, including the generation of videos with audio.

[article] What is First Principles Thinking? We can rush through life with shallow understanding of too much. This article is a good reminder to acquire a deeper knowledge of what we’re dealing with.

[blog] Why I’m not letting the juniors use GenAI for coding. Good reasoning, centered on code review behavior, learning moments, and (lack of) evolving thoughts. I don’t agree with all of this, but it’s worthwhile to noodle on.

[article] The importance of memory for AI. This was relevant to the piece above, as I do think our AI tools can “learn” from mistakes over the long term. If you have and use memory effectively.

[article] Employees brace for AI-driven change: survey. Good idea. Even if you don’t think it will affect your job (it will), it’ll affect your industry.

[blog] Mastering agentic workflows with ADK for Java: Parallel agents. I’m learning a few things from this series of posts. Move from abstract to tangible understanding of these agent patterns.

[blog] Mastering agentic workflows with ADK: Loop agents. This is a type I haven’t built myself yet, and appreciated seeing in action.

[article] Product Manager vs. Product Owner: Why Teams Get These Roles Wrong. I’ve done both jobs, and remember speaking about it a decade ago. Sometimes I think we over-think this. Having distinct roles here (versus tasks within a role) creates communication breakdowns.

[blog] The global endpoint offers improved availability for Anthropic’s Claude on Vertex AI. Instead of being pinned to a region (which you may need to do for sovereignty or perf reasons), this new global endpoint provides pay-as-you-go access to Claude wherever there is capacity.

[article] Anthropic unveils new rate limits to curb Claude Code power users. I feel for them. Amazon needed to immediately rate limit their new agentic IDE, our Gemini CLI got hammered. But yeah, users are justified in being grouchy when companies back down from prior limits.

[blog] Exploring the new BigQuery Data Engineering Agent. Get the data and clear instructions needed to go hands on with this AI-assisted data science experience.

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

Day one of vacation with my two younger kids while my wife and oldest are away at college orientation. Mini golf and such were on the agenda today.

[article] Our Favorite Management Tips on Communicating Like a Leader. Super solid advice here for anyone (everyone?) trying to get better at communicating.

[blog] The agentic experience: Is MCP the right tool for your AI future? If you’ve got a bet on MCP, you better have a strong API strategy. This post uses that lens to offer direction.

[blog] Modern Java for Cloud-Native Apps: Is It Catching Up with Go and Node.js? Not a scientific breakdown, but the dated “Java is slow” idea should be put to bed.

[blog] Your guide to taking an open model from discovery to a production-ready endpoint on Vertex AI. Fantastic post that really solidified my understanding of the workflow from selection to serving of open models.

[blog] Scaling Inference To Billions of Users And AI Agents. I want to keep seeing posts on production-grade deployment and scale, not just “getting started” content.

[blog] How AI Is Improving My Wireframing Process. I *think* AI can improve just about any process. These are practical areas where designers are going to see an uplift.

[blog] How I Turned Gemini CLI into a Multi-Agent System with Just Prompts. Fun experimentation here, with a powerful result. Being able to spawn these agents, each with a unique persona, has many practical implications.

[article] ‘Wizard of Oz’ blown up by AI for giant Sphere screen. We took the wraps off this a couple months back, and now everyone will get to enjoy the results.

[blog] The many, many, many JavaScript runtimes of the last decade. Goodness, there are a lot of runtimes for JavaScript. Check out this post for a run through.

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