↵ select ↓ ↑ navigate esc close

Seroter's Daily Reading: Daily Reading List – October 7, 2026 (#883)

Seroter's Daily Reading ·

Listen: https://blossom.buildtall.systems/410c26146e14d3b41eead3f990968802305a2c6d69b56a8511f4294b65c5bcf6.mp3

Source: https://seroter.com/2026/10/07/daily-reading-list-october-7-2026-883/


Today's post is Richard Seroter's Daily Reading List for October 7, 2026, number 883. Seroter opens with a personal note: he doesn't have much family nearby, so he's especially looking forward to dinner with his visiting brother on his birthday. He says everything in the list matters to him, but family matters much more.

He starts with The state of the tech industry in 2026, a piece by Gergely Orosz in The Pragmatic Engineer. Seroter calls it an excellent look at where things stand today, at least for software development, and adds that data, inference, and operations deserve the same treatment. The article is based on Orosz's keynote at the LDX3 engineering leadership conference in New York, drawing on visits to OpenAI and Anthropic and fresh data from GitHub, Factory AI, and Linear. The big theme is speed of change. Martin Fowler is quoted saying nothing has hit with the magnitude of AI, a whole size difference from anything before. Orosz lays out what has changed: practically nobody writes code by hand anymore, engineers commonly run five to ten parallel agents, agent-generated pull requests on GitHub grew ninefold in eight months, and by August 2026 there were more agent-authored PRs than human-authored ones. Linear reports more agent-created issues than human-created since July. The IDE is fading, with tools like Cursor, Codex, and JetBrains Air moving away from the editor model. Development often starts in Slack, companies are building their own agent harnesses, and migrations that once took years now take weeks or months. Orosz also names what is broken: assumptions about code quantity, human code reviews becoming theatrical, and quality and reliability dropping. What remains the same includes the importance of teams and planning. Seroter's point about the missing treatment of data, inference, and operations is a fair one, because the article is squarely about the software development side of the shift.

Next is Inference Is the Most Important Market in Software, by Tomasz Tunguz. Seroter asks whether inference matters more than the database category, and says that by revenue, yes. Tunguz argues that AI inference is set to surpass the database market, reaching roughly three hundred fifty billion dollars by 2027, nearly double the database forecast. In 2025 companies paid about twenty-five billion dollars to run AI models; this year the market reaches roughly one hundred thirty billion, within striking distance of Gartner's database forecast. The crossover rewires application unit economics in three ways: usage can dwarf platform fees, gross margins compress below classic SaaS norms, and customers who bring their own keys force vendors to trade revenue for margin. Tunguz compares software companies to long-haul truckers watching diesel prices surge, hunting for efficiencies with smaller models, better harnesses, and new routing techniques. Seroter's framing is useful: the database was the most important software category of the mid-2010s, and inference is now taking that role.

Then Seroter highlights Introducing Playground: Create and play custom games, from Google. His reaction is simple: amazing, genuinely, and he is excited to build some games with his kids. The post, by Maryam Karimzadehgan, introduces Playground as an experimental gaming platform that lets people create, play, and share custom games with no coding required. You describe your idea through text prompts, tweak physics, rules, characters, or environments, and test it right away in the browser. Games can be private, shared with a link, or published to a community gallery, with leaderboards and multiplayer in select genres. Google also previews an upcoming integration with Unity Spark for professional-level mechanics and high-fidelity 3D. Playground launches in the U.S. for users eighteen and older, with tiered creation access based on Google AI subscription. The connection to the broader theme is clear: creation is being opened to people who never had the technical background, and Seroter's personal excitement is a reminder that these tools are not just for work.

Seroter then links to Wake Up to Clean Branches: Automating Your Dev Routine with Antigravity 2.0 Scheduled Tasks. The article itself could not be retrieved, but from the title and Seroter's commentary we can see what it is about. It is a demo of using Antigravity 2.0 scheduled tasks to automate parts of the development routine, such as cleaning up branches. Seroter calls it a neat demo, and says he would probably only use these scheduled tasks when his machine is always on, or when he can run them cloud-side without worrying about it. That is a practical caveat: scheduled automation is only as reliable as the machine it runs on.

Next is Companies are building governed semantic layers. Most say their AI agents are getting context somewhere else, from VentureBeat. This article also could not be retrieved, but Seroter's context gives us the core finding. Direct queries to systems, along with documents, represent most of the sources called out in the survey. He asks whether it is too early for wide adoption of trusted data layers, assuming they even exist. The title suggests a gap between the governed semantic layers companies are building and where AI agents actually get their context. Seroter's skepticism is worth noting: the industry may be investing in trusted data infrastructure while agents still go around it.

Then comes Shipping is the foundation, by Sean Goedecke. Seroter's take is direct: learn how to ship, everyone needs this mindset, and everyone in tech needs the ability. Goedecke argues that shipping is the foundation under all other engineering skills, like running meetings, writing design docs, or leading teams. He draws an analogy from Dota 2, Magic: the Gathering, and Starcraft, where aggressive early strategies constrain the entire strategy space. In tech, almost everything should be compared to the default strategy of just sitting down and making the change yourself. If you cannot ship, you get dysfunctions: fifteen minutes estimating a five-minute task, un-shippable designs, bloated estimates, frustrated teammates, and trivial tasks blown into grand projects. He also addresses AI directly: shipping was never just writing code. It is about finding the shortest pragmatic path, solving little problems, and keeping managers informed. LLMs can help, but they cannot run the process end-to-end because it requires too much context about systems, people, and motivations. This is a useful counterpoint to the earlier state-of-the-industry piece: even as agents write more code, the human ability to ship still matters.

Seroter then links to Announcing MCP Toolbox Java SDK v1.0: Agentic data access for the enterprise, from Google Cloud. He says this acts as a client for your remote instance of the MCP Toolbox, and is super useful when you are bringing data into your agentic Java apps. The post, by Abirami Sukumaran and co-authors, announces the one-point-oh release of the Java SDK for MCP Toolbox. The Model Context Protocol is described as a universal interface, the USB Type-C for AI orchestration, decoupling models from data sources so developers do not have to build N times M bespoke connections. The Java SDK brings type-safe agent orchestration to enterprise Java, with transport layer abstraction, decoupled client authentication, default parameter support, and pruning of bound parameters. The post walks through a demo called Cymbal Transit, a fictitious bus network, using AlloyDB, Spring Boot, and LangChain4j to build a stateful conversational agent. The architecture separates the declarative agent interface from tool execution, and the SDK handles session tracking and secure database queries. Seroter's framing is accurate: this is about bringing data into agentic Java applications through a remote MCP Toolbox instance, and the enterprise focus is on security and production readiness.

Next is The next hurdle for AI agents: getting websites to let them in, from TechCrunch. Seroter says personal agents are suddenly getting the stiff-arm from websites, something has to give, and notes that new proposed standards are floating around. The article, by Sarah Perez, reports that personal AI agents like Meta's Muse, Instinct, and ChatGPT's Dots are being blocked by websites. Amazon began blocking Muse from its retail site, and users report issues with Walmart, airlines, Yelp, eBay, Zillow, and others. Some blocks are intentional, but many are the result of traditional anti-bot measures, like CAPTCHAs, that do not distinguish good agents from bad ones. Walmart says its blocks are not intentional and that it is partnered with Muse. Delta says it is protecting customers from unauthorized automated activity. Yelp requires agents to use its data licensing program. Cloudflare is also in the mix, with a marketplace where AI bots pay for data access. The article notes that Meta, Walmart, Stripe, Sierra, Genesys, Rocket, NiCE, and Decagon have begun working on an open standard for how AI agents communicate with businesses.

That standard is the subject of the next linked item, A New Way for Businesses and Personal Agents to Work Together, from Meta. This is the proposed standard Seroter mentioned. Meta previews the Personal Agent Protocol, an open standard being developed with Sierra and partners including Genesys, NiCE, Decagon, Rocket, Shopify, Stripe, and Walmart. The post explains how Muse browses today, applying two tests: the one honest person test, asking whether a reasonable person would do this by hand at this scale, and the if every agent did this test, asking whether the underlying system would still function. For consequential actions, Muse seeks explicit user approval. The protocol aims to make agents predictable and welcome, guide them to preferred paths, and move from browser automation to agent-to-agent communication. Meta argues that turning away a personal agent means turning away the customer behind it, and that email took off because everyone agreed on addressing, transport, and delivery. The parallel is clear: agent commerce needs shared rails. Seroter's comment that something has to give is well supported here, because the current situation leaves consumers caught between agents and websites.

Then Seroter links to Building Git infrastructure for agent-scale development, from the GitHub blog. He says few stacks were built for the type of traffic and behavior that agents bring, and this is how GitHub is evolving. The article is by Brian Celenza, a principal software engineer working on storage and core services at GitHub. The fetched text is only the author byline, so the details are thin, but the title and Seroter's context point to the same theme from the state-of-the-industry piece: agent-generated pull requests and issues are growing rapidly, and the underlying Git infrastructure has to adapt. GitHub is evolving its storage and core services to handle agent-scale development, which means different traffic patterns and different behavior than human developers.

Next is Migrating from MongoDB to Firestore Without Changing a Single Line of Code, from Google Cloud on Medium. The article could not be retrieved, but Seroter's commentary is personal: his last two app builds have depended on Firestore, and he calls it a great database. The title suggests a migration path that preserves the application code, likely through a compatibility layer or adapter that lets MongoDB-style queries work against Firestore. Without the article text, we should not infer more, but Seroter's endorsement of Firestore is clear.

Then comes Dreamforce 2026, by Rachel Stephens at RedMonk. Seroter highlights two takeaways: the personas are flattening, and user interface is not the moat. Stephens reports that Salesforce spent a good chunk of its keynote telling customers they might never need to log into Salesforce again. The company has flattened its personas, swapping the low-code to pro-code spectrum for the umbrella term Trailblazers. And it is relinquishing control of the user interface. Salesforce rebranded Headless 360 as AIforce, and Patrick Stokes argued that the product is not the UI but the trust customers put in Salesforce to hold their data, workflows, permissions, and security rules. Stephens raises sharp questions: how well does trust travel when agents escape sandboxes, and what does it mean for Salesforce to be disintermediated from its users? She notes that Slack remains an anchor, so going headless does not mean surrendering the primary UI for work. Seroter's two takeaways capture the strategic shift: the interface is replaceable, and the old persona boundaries are dissolving.

Then Seroter links to Introducing Google Cloud Modernize, transforming for (and with) AI, from Google Cloud. He says modernization got easier with AI, but there is still probably a structured flow to follow for discovery, stack selection, and monitoring the transformation. The post, by Souvik Choudhury and Tom Nikl, announces Google Cloud Modernize, an end-to-end transformation portfolio that brings together Migration Center, VMware Engine, Mainframe Modernization, and a new EKS-to-GKE Migration Agent. The central piece is Modernization Hub, an in-console experience for analyzing source code, mapping dependencies, and accelerating modernization for Java, .NET, and mainframe applications. The portfolio includes agentic infrastructure assessment with a Quick Estimator, platform modernization with new compute options like the X5 Series for SAP and M4N for databases, and application modernization with the App Modernization CLI. The post cites NetEase Games cutting server costs by forty percent and Intesa Sanpaolo using Dual Run to build confidence for mainframe migration. Seroter's point about a structured flow is reflected in the portfolio's phases: assess, migrate, and refactor, with human-in-the-loop gates and monitoring.

Finally, Seroter links to 12 things I miss since coding died, by Anton Zaides. Seroter says this resonated with him, at least most of the list, and that it is okay to reflect and admit we are losing something with more automation, while also gaining other things. Zaides writes that he has not written a line of code in more than a year, and he misses it. His list includes Sherlock-Holmesing an incident, the satisfaction of a great code review, rabbit-holing and climbing out, exploring uncharted territory, dancing with the debugger, no good break moments, automating completely unreasonable things, providing knowledge to the next generation, my-brain-is-awesome moments, shared debugging, working on a complex technical design, and when something impossible finally works. The emotional core is that these experiences will not happen again. He closes with William Bridges' framework of transitions: the ending, the neutral zone, and the new beginning. All of us in software are in the neutral zone, still figuring things out. Seroter's response is measured: it is okay to acknowledge the loss, and also to recognize what is being gained.

Across the list, a few threads stand out. One is the scale of the AI shift, from the state of the industry to inference economics to agent-scale Git infrastructure. Another is the tension between automation and human craft, captured in Shipping is the foundation and 12 things I miss since coding died. And a third is the renegotiation of interfaces and trust, from Playground opening game creation to everyone, to Salesforce going headless, to websites deciding whether to let agents in. Seroter's list holds both the excitement and the unease of this moment.

  1. Daily Reading List – October 7, 2026 (#883)
  2. The state of the tech industry in 2026
  3. Inference Is the Most Important Market in Software
  4. Introducing Playground: Create and play custom games
  5. Wake Up to Clean Branches: Automating Your Dev Routine with Antigravity 2.0 Scheduled Tasks
  6. Companies are building governed semantic layers. Most say their AI agents are getting context somewhere else
  7. Shipping is the foundation
  8. Announcing MCP Toolbox Java SDK v1.0: Agentic data access for the enterprise
  9. The next hurdle for AI agents: getting websites to let them in
  10. new proposed standards
  11. Building Git infrastructure for agent-scale development
  12. Migrating from MongoDB to Firestore Without Changing a Single Line of Code
  13. Dreamforce 2026
  14. Introducing Google Cloud Modernize, transforming for (and with) AI
  15. 12 things I miss since coding died