Seroter's Daily Reading — #877 (September 29, 2026) · Seroter's Daily Reading

Listen: https://blossom.buildtall.systems/92754f1d12a53866c5a481ce98e38832ebd321234578bd96a5b07dbff0f53265.mp3

Source: https://seroter.com/2026/09/29/daily-reading-list-september-29-2026-877/

Seroter's Daily Reading, episode 877, for September 29, 2026. In his opening remarks, Richard Seroter observes that many of the day's items encourage us to do more thinking: ask more questions, consider architectures before unleashing agents, and start from first principles. Hard to disagree with any of that.

First up is "You should all be asking way more questions," by Sean Goedecke. The piece argues that asking frequent, short confirmation questions during explanations prevents small misunderstandings from ballooning into bigger ones. Goedecke says he asks roughly a question every thirty seconds, and he encourages readers to build the plan in their head as they listen, visualizing data flow, service communication, and persistence. That habit matters even more with AI agents, which make design mistakes all the time. Seroter's take is that the advice applies to virtually every aspect of life: unlock your inner four-year-old, ask questions, express interest, and pay attention to answers.

Next is "How I converted GenLatte to fullstack Dart," from the Flutter blog, written by Craig Labenz. The article shows how a real app was migrated from a Node.js backend to fullstack Dart with the help of Gemini 3.6 Flash. The author explains that a one-to-one rewrite wasn't worthwhile; instead he redesigned the architecture to share models, shrink fifteen Cloud Run services into one Firebase function, remove client-side writes and triggers, and add end-to-end tests. The result was a much smaller server bill, plus better reliability and performance. Seroter calls this a good example of zooming out and thinking architecturally before setting AI loose on a refactoring.

Then, "Survey: Lack of Confidence in Software Supply Chain Security Runs High," by Mike Vizard on DevOps.com, covers a Cloudsmith survey of four hundred platform and security engineers. Nearly three quarters are only moderately confident or not confident in their ability to prevent supply chain attacks. Top concerns include AI-generated code introducing malicious dependencies, automated systems modifying software at scale, and attacks blending into normal DevOps activity. Seroter says he would have a lot of questions for anyone who claimed to be super confident right now, because many dependencies are being selected by AI agents rather than humans.

"First Principles Thinking," by Sunil Sadasivan, connects senior engineering flow to stepping back and understanding the basic elements of a problem. The author writes about setting long-held experience aside, the agentic era's faster learning loops, and the value of momentum and small steps. Seroter's commentary is direct: understand the basic elements and build up solutions from there. That is still something humans excel at.

"Shopify opens checkout to browser-based AI agents," from TechCrunch, reports that Shopify has expanded WebMCP support to the checkout flow, including Shop Pay. With new tools called get_checkout, update_checkout, and complete_checkout, browser-based agents can inspect a checkout, change the address or delivery option, and place an order after buyer authorization. This contrasts with retailers like Amazon and Adidas that are blocking agents. Seroter warns that if agents cannot complete transactions with your app, you are already behind.

Next is "Should workers be paid for AI skills?" from HR Dive. Payscale found that while nearly a quarter of employers treat AI fluency as an unrewarded baseline skill, most employees disagree. Because more than four in ten companies cannot find the AI talent they need, new hires can command twenty to forty percent premiums, which Payscale calls a potential retention time bomb for current employees who upskilled without a raise. Seroter frames it as an interesting problem: might you start paying new hires more for the AI skills you need, without compensating current staff who reached the same level?

"Pub/Sub AI Bytes: Part 3 — Generate and edit SMTs with Gemini in seconds," from Google Cloud on Medium, could not be retrieved in full. But the title signals that it demonstrates using Gemini to quickly generate and edit Pub/Sub message transformations, called SMTs. Seroter highlights a notable example from the piece: many times AI is not the point. The point is creating something that might have required specialized knowledge before. In this case, AI is helping users create message transformations faster.

"Why faster AI coding can mean harder engineering," on InfoWorld, discusses an increasingly common paradox. The author points to Simon Willison's observation that coding agents make software engineering harder because getting the most from them requires extraordinary discipline and knowledge. Faster implementation does not automatically create human capacity; teams may take on more ambitious projects and spend saved time on compatibility, migration, and review decisions. Seroter's take: okay, maybe faster is not always better. These new gains mean new responsibilities.

"Companies are paying LLMs to generate text for decisions that only need a label. Jev offers a cheaper way," from VentureBeat, was not retrievable, but the headline alone makes the point. Seroter says the smart insight here is that classification models are back because pre-training quality improved. Many decisions only need a label, not generated text, and cheaper classification approaches can handle them.

"OpenAI takes on Microsoft with the launch of what feels a whole lot like ChatGPT’s own office suite," from TechCrunch, covers a busy Dev Day. OpenAI introduced Space, a shared workspace inside ChatGPT; Pages, a word processor built for human and agent collaboration; and collaborative slides as an answer to PowerPoint. These tools put OpenAI more directly into competition with Microsoft, even as Microsoft and Salesforce race to add AI features of their own. Seroter notes it was a busy day from OpenAI, with a ton of Dev Day announcements, and points to OpenAI's own recap.

Finally, "Top Gen AI Frameworks for Go in 2026: A Hands-On Comparison" on xavidop.me compares five frameworks. The article finds that Go has become the language of AI infrastructure, and the choice is no longer Go or Python but which Go framework. It covers Genkit Go, Eino, Google ADK Go, tRPC-Agent-Go, and the now largely unmaintained LangChainGo, comparing developer experience, abstraction levels, observability, provider support, and maintenance. Seroter says Go has a rich ecosystem of AI frameworks at this point, and this article will help you choose.

Across the day, a common thread is clear: agents and AI can accelerate work, but only when humans supply the questions, the architecture, and the first-principles understanding. Faster output creates new responsibilities, and organizations need to think carefully about compensation, security, and where the gains actually go.

Your signing session ended