Category: Daily Reading List

Today’s list definitely has some assertive opinions. What’s the new baseline for performance? How are you thinking about MCP wrong? What’s going to happen when you’ve lost comprehension of your codebase? We should ask hard questions and noodle on the answers.

[blog] 10x is the new floor. As our tools get better, the floor goes up. More is expected. Being “just ok” at your job is a fairly risky proposition in 2026.

[blog] Introducing “vibe design” with Stitch. This is such a game-changer for UX folks, but also everyone else who wants to bring smart design into their apps.

[article] How coding agents work. Another good one from Simon that explains the agentic loops and techniques you find in coding agents.

[blog] Gemini API tooling updates: context circulation, tool combos and Maps grounding for Gemini 3. Good quality of life update for people buildings AI apps and agents.

[blog] Our latest investment in open source security for the AI era. A handful of us are pitching in to ensure that open source stays stable and secure.

[blog] MCP Isn’t Dead You Just Aren’t the Target Audience. Allen makes the important point that not every agent has a shell or is a coding assistant. For many agents, MCP is an important connector.

[article] Agents write code. They don’t do software engineering. I mostly agree with this. Today. But the line keeps moving, and if you think only humans will do engineering, I think you’ll be left behind.

[article] How Uber Engineers Use AI Agents. These engineers use AI for assigned work. Here are insights from a recent talk by one of their leaders.

[article] OpenClaw can bypass your EDR, DLP and IAM without triggering a single alert. Yes, agents aren’t ready for unfettered access to everything to do anything. But that may not last long. NVIDIA is doing work around this.

[blog] From Ideation to Automation: The Scoop on Outages. McDonalds gets grief for offline ice cream machines, but apparently there’s more going on than I thought. And, better solutions to get back online.

[blog] TikTok reduces code size by 58% and improves app performance for new features with Jetpack Compose. The right framework can make a meaningful difference in performance and maintenance cost.

[blog] Comprehension Debt – the hidden cost of AI generated code. It’s your job to understand your code and how your system works. Are you piling up comprehension debt, or building in the right discipline?

[article] Markdown is now a first-class coding language: Deal with it. There are so many ways nowadays to start nerd fights. Saying this is one of them.

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

I need the right mix of meetings in my workday for it to be a good day. Some quality 1:1 chats, a few decision-focused meetings are fine, and then seeing something that’s exciting for our users. Today was a good day.

[blog] Bringing the power of Personal Intelligence to more people. You’ll now see this in Search, the Gemini app, and Gemini in Chrome.

[blog] Subagents. Simon’s been adding to this series of posts about agentic engineering patterns, and this one on subagents is helpful.

[blog] Giving you more transparency and control over your Gemini API costs. This is a harder problem than you might think. Glad to see this team giving developers spending caps and other tools to control cost.

[article] Google Workspace’s New AI Features Seem Genuinely Useful. Nice to hear. We’re all shown a lot of AI tools and probably only use a few.

[report] The State of AI in the Enterprise. I’m surprised that this Deloitte report is ungated. Check it out for some useful information about enterprise approaches to AI.

[blog] Measuring progress toward AGI: A cognitive framework. This links to a paper where we look at 10 cognitive abilities and how you’d evaluate progress towards Artificial General Intelligence.

[article] Banks struggle to scale AI as legacy tech devours IT budgets. Until you get some of the prereqs under control, it’s going to be hard to throw important dollars at AI work. But results need to be there too!

[blog] Introducing multi-cluster GKE Inference Gateway: Scale AI workloads around the world. Run inference workloads across clusters, and even across regions.

[blog] State of Open Source on Hugging Face: Spring 2026. A metric ton of data here from Hugging Face. Which open models are used where, who is contributing the most, and much more.

[blog] Developer Guide: Nano Banana 2 with the Gemini Interactions API. It’s an underrated API and Philipp is inspiring me to make this a bigger part of my toolbox.

[blog] Agent Protocols — MCP, A2A, A2UI, AG-UI. Get familiar with these, or at least the use cases they purport to help.

[blog] Announcing the Colab MCP Server: Connect Any AI Agent to Google Colab. Wicked. Offload to the Colab host and use notebooks as tools thanks to this new open MCP server.

[docs] Durable AI agent with Gemini and Temporal. Want to persist the steps of an agentic loop so that you can resume in any situation? That’s what Temporal does.

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

I waded a bit into the “MCP or not” debate by running some experiments to see how much MCP costs my custom-built agent. If you complement with agent skills, the answer is “not too much.”

[blog] Become Builders, Not Coders. This is more of a directive versus suggestion at this point. What has to change and how do you do it? Here’s a post with advice.

[blog] Balancing AI tensions: Moving from AI adoption to effective SDLC use. The DORA team used some fresh research to understand how teams are using AI, where they get value, and stumble. The suggestions are very good.

[blog] Why context is the missing link in AI data security. These Google Cloud tools are really impressive at identifying and masking sensitive info. Now, with better context classifiers.

[blog] Run Karpathy’s autoresearch on a Google serverless stack for $2/hour. With the exception of doing massive training jobs, most of us can try out nearly anything with AI for a reasonable cost. I like Karl’s example here.

[article] Why the World Still Runs on SAP. Big ERP, CRM, and service management platforms aren’t going anywhere. But it’s going to get easier to set them up, use them, and operate them.

[article] You’re Not Paid to Write Code. I recognize that I’ve shared a lot of posts on this topic. But it’s important. We’re not just adding tools to the mix; we’re changing identities and habits. That takes repetitive reminders and motivation.

[blog] When to use WebMCP and MCP. Pay attention to WebMCP. It might turn out to be something fairly important.

[blog] BigQuery Studio is more useful than ever, with enhanced Gemini assistant. I like this surface, and it’s made data analytics so much simpler for experts and novices.

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

A little throwaway tweet yesterday somehow turned into my most viral thing, maybe ever. I don’t understand social media. But it was also an awesome reminder that many people have no idea what the AI on their phones, email client, and corporate systems already does!

[blog] A2A Protocol Ships v1.0: Production-Ready Standard for Agent-to-Agent Communication. Congrats to the team here. A few things got better in this release, and I expect it to continue getting adopted within products and by developers.

[blog] BigQuery pipe syntax by example. Lots of examples here, and you can try out this SQL alternative in our free BigQuery sandbox.

[blog] How to Do Code Reviews in the Agentic Era. I liked this take. If you’re in OSS, you already have a zero-trust approach to contributions. Who cares where the code comes from? This is what Daniela is looking for.

[article] WTF does a product manager do? (and why engineers should care). Good post. What a PM does hasn’t changed a ton, but the way they do it has. Or at least should!

[article] Preparing your team for the agentic software development life cycle. In my little bubble (regarding what customers constantly ask me about), this is the #1 topic.

[blog] Right-Sizing Engineering Teams for AI. Some quick thoughts that are worth checking out. What’s the ideal makeup for an engineering team in 2026 and beyond?

[article] How is AI already reshaping the software engineering labor market? Let’s stay on this topic, I guess? More advice for tech team leaders.

[article] What Authentic Leadership Looks Like Under Pressure. This feels related to the preview three pieces. This is likely a stressful time for many of us. How are you leading in this moment?

[blog] MCP vs. CLI for AI-native development. The “MCP or not” debate hit a fever pitch this week. Something’s in the water. It’s an “and” conversation to me; MCP makes sense in many situations, not in all.

[article] The case for running AI agents on Markdown files instead of MCP servers. Now we’re talking about skills versus MCP. Again to me, the answer will be “both” for a lot of cases. I’ve been testing this out myself.

[blog] Twenty years of Amazon S3 and building what’s next. Feels like this is what started the mainstream cloud story. Congrats to the Amazon team on 20 great years.

[article] What OpenClaw Reveals About the Next Phase of AI Agents. We see time and time again that you shouldn’t dismiss an early, rough introduction of a new technology. It often signals that there’s fresh appetite for an unmet need.

[article] NanoClaw and Docker partner to make sandboxes the safest way for enterprises to deploy AI agents. Safety features always follow a buzzy new idea. Just wait a bit and things like this pop up. More here.

[blog] Simplify your Cloud Run security with Identity Aware Proxy (IAP). Fantastic feature for people who want authenticated web apps with as little work as possible.

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

It was a day. But we had a fun read-through of our Google Cloud Next developer keynote. I’m excited to see many of you in person soon!

[article] AI productivity gains are 10%, not 10x. We’ve said the same thing publicly. There are tasks that have 3x or 10x productivity gains, but it’s not uniform across the whole day or entire value stream.

[article] CEOs think AI use is mandatory — but employees don’t agree, survey says. This story posts about the disconnect between execs and employees, but notice the blurb about middle managers. You don’t win that tier over, every initiative tends to die.

[article] Pity the developers who resist agentic coding. I wouldn’t use the word pity at this point. This article points out that devs are missing the thrill of really building at the speed of thought.

[blog] Cloud CISO Perspectives: New Threat Horizons report highlights current cloud threats. Even if the threats themselves don’t change (they do), notice how bad actors seamlessly switch to the ones getting less attention.

[blog] Introducing Replit Agent 4: Built for Creativity. How we work changed. Stop fighting it. Tools like Replit do a great job of showing what the future looks like.

[blog] What you need to know about the Gemini Embedding 2 model. This is new, but not getting the attention it deserves. This new embeddings model makes life much easier for those with a mix of data.

[blog] Human Insight, Amplified: How Forrester Is Reinventing Research For The AI Era. This seems like a good idea. Analyst firms need to rethink their research approach, and distribution. This addresses the latter.

[blog] Protecting cities with AI-driven flash flood forecasting. Great work from Microsoft Research to make this capability available to local communities.

[blog[ 5 design skills to sharpen in the AI era. From Figma, this seems like a useful list of areas to focus on.

[blog] Inference on GKE Private Clusters. Good use case! Can a Kubernetes cluster with no internet access still do AI inference? Yes, yes it can.

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

Sometimes a different model really does help. I was building an agent last night for fun, and couldn’t get it to do what I wanted. I upgraded to the latest Gemini model and now it works like a charm. That was the only change!

[blog] Welcoming Wiz to Google Cloud: Redefining security for the AI era. I’m excited that we closed this acquisition and now have this talented team with their differentiated security platform. I’m not sure anyone has a security stack quite like ours.

[article] Andrej Karpathy’s new open source ‘autoresearch’ lets you run hundreds of AI experiments a night — with revolutionary implications. Besides coding (which I don’t get to do every day), my main use of AI is for research. Autoresearch could absolutely be a transformative thing.

[blog] The 8 Levels of Agentic Engineering. Fantastic post that articulates each progressive stage of using AI in engineering, and what you gain from each.

[blog] Plan mode is now available in Gemini CLI. Great functionality added to the Gemini CLI. Do safe, read-only mode work before jumping into action.

[blog] Cost-Effective AI with Ollama, GKE GPU Sharing, and vCluster. As I see people struggle with token costs or model availability, I’m definitely of the mind that there’s clear cases where you want to run models yourself.

[blog] When Developer Workflow Discipline Isn’t Enough. It’s about platforms, and how you’re serving AI functionality at scale in large companies.

[article] 10 Hacks Every NotebookLM User Should Know. Great list of ways to personalize the experience and learn on your terms.

[blog] From games to biology and beyond: 10 years of AlphaGo’s impact. This was a turning point for AI, and we may look back at this as a pivotal moment.

[article] How to Quash Your Fear of Messing Up. What causes you to hesitate? How can we think about risk differently? Read this for advice.

[blog] Best Practices for Secure Error Handling in Go. Even if you don’t write code in Go, you’ll learn a few useful things about avoiding security issues with your error handling.

[article] Layoffs, cost-cutting shatters IT worker confidence. Understandable! Tech workers faces a lot of simultaneous stresses.

[blog] Bring Your Database Tools to the Agent Skill Ecosystem. Very cool new capability to turn MCP toolsets into Agent Skills. Worth trying this out.

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

We shipped some sweet AI updates today—AI app framework, unique embedding model, smarter AI in Workspace—but I read through more than just that 🙂

[blog] Gemini Embedding 2: Our first natively multimodal embedding model. Wow. Map text, images, video, audio, and documents into as ingle embedding space. Unique, and powerful.

[blog] Gemini in Google Sheets just achieved state-of-the-art performance. We can all become spreadsheet masters now. Check out what’s now in your broader Google Workspace toolbox. I should use more of this!

[article] Dynamic UI for dynamic AI: Inside the emerging A2UI model. This is a good dive into the new frontend paradigms to think about now, and how A2UI works.

[blog] Your Data Is Made Powerful By Context (so stop destroying it already). You’re screwing up key data needed by your AI systems by separating observability signals in different “pillars”, says Charity.

[blog] Extend your coding agent with .NET Skills. Cool. This should become standard son as languages provide out -of-the-box skills to use them correctly.

[blog] Announcing Genkit Dart: Build full-stack AI apps with Dart and Flutter. I find this VERY intriguing. Model agnostic, run anywhere, and built-in AI flows for your apps.

[blog] Cracking the code on corporate visibility. You’re doing some great work. How come you’re not getting the right credit for it? Consider being more visible by creating and sharing content.

[article] A2A vs MCP – How These AI Agent Protocols Actually Differ. Impressive level of detail in this new tutorial from DigitalOcean.

[blog] Fixing AI Slop with a Skill in Gemini CLI. I don’t get super riled up if I know I’m reading AI generated text. What I want is text that sounds like a human. This skill fixes the default mode of text generation.

[article] AI coding assistants may influence which languages developers use. I could see that. I’m functional in four programming languages, but I tend to generate AI code in only one or two of them.

[article] The “Last Mile” Problem Slowing AI Transformation. Even enthusiastic adopters of AI tools hit issues. Where do they struggle to get through the last mile, and how can we all learn from them? Some tips here.

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

I had a fun weekend, with no option to do anything work related. Baseball, boating, and shenanigans with family. I probably get more inspiration for things to do at work because I’m living an enjoyable life outside of it. At least that’s what I tell myself.

[blog] Designing MCP tools for agents: Lessons from building Datadog’s MCP server. Some hard-earned lessons here! There are at least 3-4 strong pieces of advice in this Datadog post.

[article] The revenge of SQL: How a 50-year-old language reinvents itself. Is SQL hot again? Is it about relational databases solving most use cases nowadays? Also you have SQL on the frontend, better SQL clients and more.

[article] EY hit 4x coding productivity by connecting AI agents to engineering standards. Better models matter. But applying a smart context is a difference maker regardless of model.

[blog] Game on with Spanner: How Playstation achieves global scale with 91% less storage, 50% lower costs. Cool story. A high performing database engine can end up saving you a ton of money and complexity.

[blog] How Do Large Companies Manage CI/CD at Scale? Me building and running a simple deployment pipeline is not “scalable CI/CD.” What do teams do when they have lots of apps, pipelines, and targets? Some insight here.

[blog] Go for Backend Development — Why We Bet on It. Very strong, defensible case for using Go.

[article] When Using AI Leads to “Brain Fry.” Across roles, people are using AI past the point of their brains can handle. What leads to brain fry, and how to prevent it?

[blog] Hardware-Enabled Software and the Next Generation of Vertical AI. I don’t pay a lot of attention to this space, so this was educational.

[blog] Firebase A/B Testing is now available for the web. The functional was available to mobile devs for a while, and now web users can take advantage of this powerful system for running experiments.

[blog] Terminals Are Cool Again. Maybe we should be building more terminal apps? I’m not sold that they’re more accessible than a web or desktop app. But there’s no doubt they’re lighter weight and can be more efficient to use.

[blog] gRPC on GKE for Fun & Profit Part 1 — An Overview. gPRC is a key technology within Google, and also many other companies that care about performance between services. See part 2 of this series as well.

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

I’m off to Arizona for a couple of days to watch Spring Training baseball with my son. And to hang out with my brother and friends. Back here on Monday!

[blog] You can’t stream the energy: A developer’s guide to Google Cloud Next ’26 in Vegas. If you’re procrastinating, stop it. I saw the numbers this week and the event is close to selling out. Get yourself to the premium dev and AI event of the year.

[blog] Vibe Coding to Production. Even now, you’re probably not pushing production apps from your IDE. Ankur connects his AI-built app to GitHub, Cloud Build, and Cloud Run.

[blog] Does AI Make Us Smarter or Dumber? Yes? We’re losing some “primitive” abilities but unlocking new superpowers.

[article] OpenAI launches GPT-5.4 with Pro and Thinking versions. Plenty of new models this week, including fresh ones from OpenAI.

[blog] Look What You Made Us Patch: 2025 Zero-Days in Review. The zero-day landscape was different last year. Even more enterprise tech attacks, with browser-based exploitation dropping.

[blog] How Google Does It: Applying SRE to cybersecurity. SRE applies to security too, of course. I like these details of how we think about it.

[article] The Pulse: Cloudflare rewrites Next.js as AI rewrites commercial open source. I guess we can just rewrite stuff now? It’s not difficult to regenerate entire projects, build compatibility layers, etc.

[blog] Can coding agents relicense open source through a “clean room” implementation of code? Continuing that thought, how legal is it? That question is being tested here as debate arises over a rebuild and relicense.

[blog] GitOps architecture, patterns and anti-patterns. Are you following good practices, or anti-patterns? A lot of specifics here.

[article] Cursor is rolling out a new kind of agentic coding tool. Looks cool. Always-on fleets of agents will be a commonplace thing in twelve months. Maybe six.

[blog] Practical Guide to Evaluating and Testing Agent Skills. You know how to build the skill, but can you test the skill? It’s really the most important part, the testing. Anybody can just build them. Yes, I’m ripping off a Seinfeld bit.

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

Today got away from me. I just looked up and it’s 5pm. I’m taking a half-day tomorrow to head out on a short vacation, so expect an earlier-than-usual reading list publication.

[article] Google ADK Opens the Door to AI Agents That Work Inside Your DevOps Toolchain. When you can build agents that easily connect to your most important dev tooling, you open up a lot of possibilities for automation.

[article] You Should Take That “Boring” Meeting. Apparently we underestimate how interesting a “boring” meeting can be. Maybe?

[article] An ode to craftsmanship in software development. I get you. Maybe that beautifully designed algorithm or flawless project structure was never the art of software.

[blog] Grow your own way: Introducing native support for custom metrics in GKE. Autoscaling is amazing, but you should be able to trigger it based on your own metrics. Now you can.

[blog] Three AI Blind Spots That Cost You Developers. No one will find our use your product correctly if you describe it using weird corporate talk. Are you writing for humans and agents right now?

[article] Does using AI to code come at the cost of learning? Good post. We’re not doing deep learning if we rush through an exercise with AI doing it all. Maybe that’s fine for a subset of tasks, but recognize when you need to really absorb something.

[blog] Integrate NotebookLM with Gemini CLI, Google Antigravity or Other Agents with MCP. Creative use of private APIs! I don’t think we officially support this scenario, but life finds a way.

[blog] A Monorepo Is NOT a Monolith. If you’re using a monorepo for all your code, does that mean you have a monolithic, tightly-coupled system? It does not.

[article] Does AGENTS.md Actually Help Coding Agents? Apparently LLM-generated instructions don’t help nearly as much as human-written ones.

[blog] The ultimate Nano Banana prompting guide. These are great tips for how to structure prompts to our image model.

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