Category: Daily Reading List

Issue 900. It’s only a number, but I like big round numbers. If you’re in America, enjoy your Halloween and keep the best candy for yourself. The kids can have the abandoned Twizzlers.

[blog] Antigravity Plugin Marketplace in Action: Auditing My Talk Slides with the Google Slides Plugin. Excellent use case. Use agents and plugins to keep content up to date.

[blog] Announcing Project ProtoStar: Making Publicly Funded Science AI-Ready. Very cool. The US invests a ton in research but the output isn’t always usable by everyone. This looks like it will help.

[article] InfoQ AI, ML and Data Engineering Trends Report – 2026. The “innovator” and “early adopter” side of this report is stuffed with areas to investigate and apply.

[blog] Hill-Climbing Agent Quality with Agents CLI Evaluation. Really good post that explains hill-climbing with an agent, and what we’ve learned about each step.

[article] “I don’t think the world needs a new GitHub”: What GitHub’s leaders really think about their new rivals. I don’t think so either. But GitHub has to evolve, and they know it. Rooting for them.

[blog] Prefix Caching and the Hidden Cost of Editing Chat History. This was new information for me. Keep the start of your prompt static!

[blog] Innovate on your own business terms: Google Cloud spend caps now generally available. I missed these features when reading the release notes a few weeks ago. Now you can set spending caps via API, view a real-time dashboard, and count on automatic resets each month.

[blog] How I Cut API Response Times by 60% with Go Caching Strategies. Stuff like this feels like knowledge, and something you should have familiarity with before vaguely asking an AI agent to “implement caching.”

[blog] Why are good developers so wary of agentic development? Going all-in on a technology/paradigm is a big deal, especially when you’re already experienced in that field.

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

The day started with a customer session at 6am, then onto GitHub Universe for a main stage panel discussion, follow-on Q&A, and a press interview. And now flying back home.

[blog] Designing Agentic Interfaces with A2UI and Genkit. If you’ve been reading this daily list for any amount of time, you probably know that I dig generative UIs right now. Daniela puts it through the paces with Go.

[article] What happens when the agent is long gone, but its access key lives on? Good question. I hadn’t really thought of it. But it seems important to have a strategy for this.

[blog] Behind the scenes, two months later: How we ship Google Agent Skills like products. One thing about Google, we look at everything as an engineering problem. For agent skills, my team built a robust strategy and set of platforms to generate, support, and eval them.

[article] You can’t find 10x developers, but you can build 10x organizations. There’s no doubt that many individuals are now remarkably productive. But software usually happens in teams. Activate the group and its workflow, and you win.

[blog] Coding agents need guidance, not faith. Good input, relevant context, and smart guardrails. Without that, you’re just fooling around.

[blog] How not to run an AI code review. Serve the elephant whole. We’re all figuring it out as we go along. These are some important lessons.

[blog] Gemini Robotics Embodied Reasoning 2 is now available on Google Cloud. Use this unique model with the protections and management of Google Cloud supporting it.

[blog] Getting good at AI takes time and practice. Nothing earth-shattering here, but a helpful reminder that using AI is a skill of its own.

[blog] You can now generate, preview, and edit documents, PDFs, presentations, spreadsheets, and more directly in Gemini. We’re late in adding this, but I’m happy it’s there now.

[article] AI agent access is moving faster than identity controls can govern. Seeing an uptick in this type of content. That’s probably good, as we need to really figure out what production-grade agents look like.

[blog] Design Tenets for Agent Identity Management Systems. I haven’t seen a lot of NIST content lately, but this is useful and timely. Check this out for a checklist of what you should be doing to properly manage agent identities.

[blog] Near-zero downtime upgrades: Announcing Cloud SQL for MySQL blue-green deployments. Take some risk out of database upgrades. I like that there are no app changes, and that it’s automated with near-zero downtime.

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

I’m heading up to San Francisco tonight to attend GitHub Universe tomorrow. First time! I’ll be part of a panel discussion, and then fly home tomorrow night. Looking forward to meeting a few people face-to-face for the first time.

[blog] Gemini 4 Carbon: our next frontier model. It’s us again. Two weeks later, another frontier model release. This one builds on the last, and shows great improvement at long-horizon tasks. More here.

[blog] Q3 2026 earnings call: Remarks from our CEO. Big week of public company earnings. Google had a strong quarter, and Cloud is an absolute rocket ship.

[blog] The State of Context Engineering. Excellent read. Get smarter on context fails, patterns, and opportunities.

[article] “The panic hasn’t turned into patching”: What security researchers really think about the PHP2Shell response. The vast majority of exposed systems haven’t been patched. Are we failing our collective day-2 responsibilities here?

[blog] Apple’s git still trusts the compromised eMudhra certificates. It’s hard to patch a bunch of stuff quickly, but it seems our industry hasn’t been super fast at closing the holes identified last week.

[blog] The Web Can’t Afford a JavaScript Runtime Monoculture. Consolidation is a natural response to an upfront explosion in options. Are we landing on too few JavaScript runtimes now with Deno going away?

[blog] The sleep framework: A new way for LLMs to retain knowledge in continual learning. How can LLMs continuing learning new facts, while not forgetting what it already knew? It’s a challenge, but something we’re looking at.

[article] How do developers keep momentum as AI agents change the work? Does AI mess with your flow? Can you get and sustain momentum when agents are involved? Interesting research here.

[blog] I don’t write code anymore. Ditto. Granted, I’m not paid to do so anymore. But I’m also producing more software and having more fun doing it than any time in my career.

[blog] GitHub Universe 2026: Evolving every layer of the stack. Looks like a solid vision for software engineering work in the AI era. More here.

[blog] Fast, reliable A2UI, with swappable inference formats. Now we’re talking. This is moving from being a useful idea, to a production-ready implementation. Fast generative UIs are here.

[blog] Agentic and Flexible: A next era of Android development. Speaking of A2UI, it’s not part of the Android experience with Jetpack Compose. This, and much more announced today.

[article] Our Favorite Management Tips on Delegating Effectively. This is meant for human managers, but I read much of this advice thinking about how it applies to delegating work to AI.

[blog] What was the turning point for Postgres? It’s winning right now, but why? Definitely wasn’t an overnight success story.

[blog] How AI agents shape experts’ judgment calls. Are experts influenced more by AI than they think they are? Anthropic did some research, and it looks like people tend to adopt the agent’s direction and claim as their own.

[blog] Why “Plan Mode” is just scoping. Don’t forget discovery! Want to stop the agent before it starts writing code? You have a few options, and Alexis points out when to use which.

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

Yesterday’s list had a pile of good advice, and so does today’s list. Whether it’s doing test-driven development, measuring productivity, thinking about day-2 ops, or making your docs agent-ready, there’s something for everyone.

[blog] Dancing backwards with Go: why TDD is weird. In any language, TDD can feel weird. Writing the failing test first, and then make it pass? I’m not sure how well LLMs comprehensively do it.

[blog] How to measure AI productivity? Still a remarkably important question. You say you’re saving time with AI, but what outcomes changed? Measure the waiting, not just the individual tasks.

[blog] Tokenomics (Or the right tool for the right job). Matching token use to outcomes isn’t easy for anyone. That’s enough to drive a CFO bonkers.

[blog] From Python to Rust, Go, to… Java? Steering Your AI Coding Agent Through Architecture Trade-Offs. Digest what Alexis is showing here. Vague requests to the LLM/agent yield some default languages and architectures that may not match what you really want. Be clear with your app needs so that the “default” language matches your use case.

[blog] Day 2000 Doesn’t Care How Fast You Shipped. Truth. Is your platform helping you maintain and operate all the software you’re delivering?

[blog] 100% of your developers should be building things to improve the workflow. Are they? Is that the case with you, or the place you work? Probably not. But how do we switch our focus to the systems that produce the output?

[blog] The Git Forge for Agent-Human Collaboration. I really like where they’ve landed here. This looks like a legit alternative to traditional git solutions. More here.

[blog] Bringing Google TPUs to Oracle Cloud Infrastructure. Everybody gets a TPU. This valuable processor is getting wider access, and more tools for using it.

[blog] Keeping a software factory stable at 1,000 PRs a week. Build things to improve the workflow, and increase your trust in the system.

[blog] Are agent harnesses dying? What harness distillation changes. Today’s harnesses may be unnecessary in six months. The model will do it. But I’d imagine we’ll want something that sits outside, even for the option to swap models.

[blog] Making documentation pages agent friendly. My team’s doing a lot of work here, and Tom encourages everyone to understand how agents see their docbase.

[blog] Testing and Serving A2A Agents from the Terminal with A2A CLI. Here’s a good walkthrough of this useful tool for interacting with agents over A2A.

[blog] The push to modernize legacy tech. It’s not hard to regenerate code. But you still need a good reason, and must be aware of the risks.

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

Today’s list has a ton of interesting advice, especially on team/individual workflows.

[blog] My AI Coding Workflow, v3. I liked this post. I’m bearing on “how to use this product” skills since the models and harnesses can handle that. But I like skills that are about workflows and process preferences.

[blog] Introducing new self-learning context enrichment in Knowledge Catalog. I’ve had this “ready to read” for a few days, and should have read it sooner. Very cool functionality that I’m going to go hands-on with soon.

[blog] Building agents: When to fold subagents into your main agent loop. Whither subagents? Fold more of the routing and work into the main agent loop? Anthropic talks about it here and makes a good case.

[blog] How Google Beam connects military families globally. This hit me hard for some reason. I live in a military town, and I can see how tough it is for families to be apart.

[article] AI delivers ‘zero financial payoff’ to most companies, Apollo economist says. That doesn’t seem like good news. Profit margins and revenue per employee is flat, even with AI? From what I’ve observed, few have made the structural changes to take advantage yet.

[article] 3 Paths to Delivering Shareholder Returns with AI. Speaking of making structural and capability changes, read this. Solid advice.

[blog] Why are we still building apps just for humans? You shouldn’t be. Get that logic out of the user interface agents won’t use.

[blog] Preview, edit, and collaborate on markdown (.md) files natively in Google Docs and Drive. 10/10, no notes. Do full lifecycle work with markdown files without leaving Google Docs.

[blog] Your AI Agent Doesn’t Need to Keep Every File It Reads. Smart thinking. We’re reading a lot of context that isn’t needed once a turn is done. But all that bloat costs you something.

[article] Stop building agents that can’t change models. Could you swap models and harnesses in 48 hours? You might need to at some point. Think about what needs to be in place to not start from zero when it happens.

[blog]How Gemini 4 Argon Upleveled Agents on Antigravity: Trillion-Edge Graphs and Custom TPU Kernels. A year of work done in five hours, and a smart solution to boot? Not bad.

[blog] Don’t ask AI to write the user story yet. Get questions from your LLM (and your team, and yourself), don’t just have the AI write out requirements based on vague input.

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

That’s a wrap on the week. I’m ready for a couple of days off!

[blog] Tokens to Deploy: Agent Skills are Great but CLI’s are Better. Such a great post. Which deployment approach eats the fewest tokens? Esther looked at the viable options and brought receipts.

[article] When are developers receptive to proactive AI suggestions? It’s pretty clear that developers only want eager AI suggestions at natural boundaries of their work. Not randomly in the middle.

[article] Eight reasons coding agents are better than software developers. The writer will have the counter piece (“why developers are better than agents”) next week, so don’t freak out. But if we’re being honest with ourselves, these points are valid.

[blog] Getting Started with the Google Cloud CLI Remote MCP Server. We just saw above that CLIs are an efficient and effective tool for agents. But maybe your agent can’t call local CLIs or tools. That’s where a CLI-over-MCP comes in.

[blog] Rewrite Everything in Java: Not Rust. Well that’s a bit of cold water on “Rust all the things.” Use whatever language you want, but some are wildly overhyped right now.

[blog] AI makes it easy to migrate. Flutter makes it easy to scale. Read this. It’s a good argument for why you should be thoughtful about duplicating codebases for different platforms.

[blog] Software is slow because of bad design, not slow languages. Tell me he’s wrong. Your speed problem is not about your choice of language. It’s design.

[youtube-video] AI harness crash course. People are using the word “harness” to represent many things. Annie does a good job here explaining what a harness is about.

[blog] When to use a diffusion LLM: I compared DiffusionGemma and Gemma 4. Diffusion LLMs are cool and instances like DiffusionGemma generates entire token blocks at once. Super fast. But Karl shows that “regular” Gemma 4 is still the best choice for step by step reasoning.

[article] PHP2Shell was only the beginning: AI is accelerating the patch race. Worst week ever in cybersecurity? Possibly. It’s not going to get better any time soon.

[blog] Streaming comes to API Gateway: real-time and AI-ready APIs, without changing your spec. Here’s a great option if you have streaming results from LLMs, agents, or APIs and want a proxy in front.

[blog] Introducing the Elephant-Goldfish Model plugin: scaling design-driven AI software development. Have you heard of this? It’s a way of thinking about AI coding, and we’ve now open sourced an agent plugin that you can use too.

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

What a mess of a week around tech security. I’m counting at least a dozen vulnerabilities or leaks across every layer of the stack: Linux, source control, runtime, PaaS, even certificate authorities. Stay safe out there.

[blog] Integrating TorchTPU into TorchTitan and Optimizing Models for TPU. More work between Google and Meta to scale PyTorch on TPUs.

[article] How to Use AI to Become a More Effective Engineering Leader. In October of the year 2026, we shouldn’t HAVE to say “don’t just use AI for coding”, but here’s a good reminder that it’s useful for many other things.

[blog] 30 Months on the Exponential (in Charts). What a wild couple of years so far, and we’re just getting started. Ethan’s got a new book out, and talks about this moment.

[blog] Introducing the Gemini API CLI. This new experimental CLI is a no-dependency binary that your favorite tools can use to interact with the Gemini API.

[blog] We gave AI agents a mobile dev environment. Cool solution from the Shopify folks that gets processing off the local machine.

[blog] Powering the Agent Development Lifecycle with any framework, from ADK to LangChain and beyond. Use this helpful CLI for agents that are built with frameworks other than ADK.

[blog] 5 takeaways from the State of Software Delivery Q3 Pulse report. Coding more, but shipping at the same rate? And it looks like the leading performers are getting clear separation from the rest of the pack.

[blog] Inside Agent Compute Engine: Building the compute foundation for the agentic era. Micro-VMs and more for isolated, fast compute the right way. More here.

[article] CIOs unprepared for agentic workforce shift. Oof. Is it really only 4% of tech leaders revamping roles for human/AI scenarios?

[blog] Expanding Claude availability and deployment choice on Google Cloud. This is a cool option where you want access to day-0 Anthropic features and Anthropic terms of service, but the cloud management layer of Google Cloud.

[article] 1Password launches a way for AI agents to log in on your behalf without seeing your password. It’s been a disastrous week for security. So many breaches and vulnerabilities! This is one way to limit what your agent has knowledge of.

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

Where’s your head at on personal agents? I’m using them more at work, spinning up various tasks (“go figure out what this new API is”, “turn my notes into a presentation”). But I’m not yet doing a ton on the personal front. A few research projects. Nothing daily. You?

[blog] Google Cloud expands model choice and flexibility, bringing OpenAI models to customers. Huge deal. Only one cloud can offer OpenAI, Anthropic, and Google models. That’s us.

[article] Critical PHP vulnerabilities: POC released, attacks starting. This is a big one. Do an audit and start patching immediately. More here.

[blog] Responding to core PHP vulnerabilities CVE-2026-109618 and CVE-2026-109619. I hadn’t realized our role in finding and patching this vulnerability. Here’s more on the background, and how to patch any systems running on Google Cloud.

[blog] What is a Forward Deployed Product Manager, anyway? Hmm. This post answered a bunch of questions. I have others. The concept has some merit, but I’m skeptical of the real-world application of this.

[article] AI Benchmarks: Climbing the Wrong Hills. Fairly thorough analysis here regarding model benchmarks, and the probably with making a measure the target.

[blog] Hugging Face ❤️ TorchTPU: TPUs, Natively. The Hugging Face team wrote up a great post explaining how much easier it is to now use TPUs in their ecosystem.

[blog] Genkit Go 1.14: Typed HITL tools, budgets, context compression, GenAI tracing, and decision models. Honestly, this framework for Go is getting very good. It’s making it much easier to add agentic capabilities to apps. Give it a look.

[blog] The bridge to agent-scale Git infrastructure. This post has a bit of a dive into some of the storage approach for GitHub, and changes they made for better scale.

[blog] Stress-Testing Gemini 4 Argon: Building a 3D Flutter Kart Game in 15 Prompts with Zero Assets. You can just build 3D games. In this case, with fifteen prompts over three plus hours.

[blog] SIMD and Go. I’m not a hardware guy or smart about all the CPU architectures. But I took some extra time to really understand this, and why even regular devs should care.

[article] AI made developers faster. Is their judgment keeping up? Are engineers doing anything with AI that they couldn’t do before? And are they growing their judgement at the same time?

[blog] BigQuery ObjectRef now GA: Build multimodal data platforms for conversational analytics entirely in SQL. This is a very cool solution for storing pointers to objects, and then accessing those objects for analysis in BigQuery.

[article] Claude Duet is Anthropic’s answer to Meta Muse and OpenAI Dots. Hot space. Everybody wants to give you a personal agent.

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

Heading back to San Diego after spending today in a team offsite. It’s fun working on 2027 goals knowing we’ll completely change them again in three months.

[blog] TorchTPU is now open source: Run native PyTorch on TPUs. Big news for PyTorch users. Alongside CPUs and GPUs, TPUs are now a native option.

[blog] How the Claude Code team does code review. Yes, they still do code reviews. No, people aren’t very involved. But if you learn from the AI mistakes and codify fixes into the process, you end up with a very good system.

[blog] Speeding up code review with ‘small-change.’ Let’s stick with code reviews. This post explains a system that routes reviews for human review if needed, and which get very fast approval.

[blog] No more tech debt. Wasn’t this phrase in that John Lennon song “Imagine”? I’m misremembering. Now you can just rewrite versus handle all the quirks of an existing codebase.

[blog] Bringing the borderless Lakehouse into the agentic era with innovations for Gemini. What are the chances that ALL your data lives in a single place? Zero? Being able to (virtually) pull it together is about to be even more important than before.

[blog] How Software Trusts Software: MCP, A2A, & the Registries Agents Use. Protocols like MCP and A2A help AI agents find tools, but they only verify who published them, not whether they’re safe. Registries play an increasingly-important role.

[repo] secure-tdd-agent-framework. My colleague Aron has a repo with some excellent skills for a secure development process. Take a look.

[blog] Announcing llm-d 1.0: Open distributed inference for agents on Google Cloud TPUs and GPUs. This open serving stack for Kubernetes is now ready for prime time.

[article] Training Your Own Models is an Incredibly Stupid Idea. Probably right. You don’t need to bake your data into the weights to get value from AI.

[blog] Developing Taste. This doesn’t come quickly or easily. That’s ok. But surround yourself with the right people and ask the right questions.

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

I’m in Sunnyvale today at an Agentic AI Executive Forum where Google is hosting a couple dozen open source contributors across a wide range of companies. I’ve learned a lot!

[blog] Agentic Verification. Yes, now the hard part is figuring out if what we get from these AI tools is what we wanted. Addy looks at a few verification approaches.

[article] Turn Vibe-Coded Prototypes Into Tools Your Company Can Actually Use. Skipping a SaaS product renewal because you can build it yourself? Maybe you can. But there are ways to ensure you’re building something that’s more than a toy.

[blog] The Thin Orchestrator: How We Made Our AI Agents More Reliable by Giving the Router Less to Do. I liked this post from Meta engineering. It has some very good advice for agent architectures.

[article] Code Review After Code Gets Cheap. Back to verification again. Small changes can have a huge blast radius. Don’t blindly trust the AI output.

[blog] Beyond Chatbots: How to Schedule Autonomous ADK Agents on GCP with Cloud Scheduler. How are you starting agents? Human triggers? Event-driven moments? Time-based scheduling? This looks at the latter option.

[blog] Personal software factories are a waste of time. Going overboard building a customized software factory for all your development work? Chill out, says Peter.

[blog] Modernizing perimeter security: Scalable perimeters with VPC Service Controls. Is your network perimeter set up to handle all the new types of workloads and interaction patterns? This seems like a good set of changes.

[blog] Stacked PRs: what they are and how to run CI on a stack. Heard about stacked PRs, but not sure what they really are or how to use them?

[blog] What requirements don’t you have? Real strategy is about what you’re NOT doing, and some of the best architects know what NOT to require of their app. LLMs aren’t great at this, and need human oversight.

[blog] Step by Step guide: Build a Voice Agent That Never Goes Silent using ADK. Good stuff. I want to get more hands-on (agent-on?) with this pattern of async tooling.

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