Dear reader,

Greetings from the Buttondown world!

The change was a deliberate choice. At first, I wanted to build my own newsletter because I love Django. But as with anything, building is only half the work. The bigger task is maintaining it. I am proud to share that Buttondown is built with Django, so it stays consistent with my website, and the newsletter is officially under my domain name too.

What I like most is that you, dear reader, can comment on my work without creating an account. And it motivates me to share my endless life and work stories with you.

In other news, I hope your week is going well and that you are floating with the sky lanterns.

Build on ECS Primitives, Not Convenience Layers: It is safer to build critical systems on durable infrastructure primitives such as ECS, EC2, S3, and IAM rather than higher-level convenience layers. Wrappers and abstractions should be evaluated carefully for lock-in, long-term support, adoption, and whether the underlying infrastructure can continue to operate if the abstraction is discontinued.

I’m Starting to Think Every Developer Should Build Their Own Internal Tools: Developers learn more by building internal tools that scratch their own itch repeatedly rather than creating practice projects. Being both the user and the builder creates a loop that improves the tool, saves time, and streamlines the workflow.

Cognitive Debt: The Hidden Cost of the Agentic SDLC: AI-assisted programming can create “cognitive debt” when teams ship code faster than they can review and understand it. We should treat the review and understanding as the real work. Keep diffs small, make developers explain their own changes, document the reasoning behind decisions, and spend AI-generated productivity gains on testing rather than simply producing more features.

The birth of the Software Verification Engineer: As AI agents generate more code, engineers are evolving into verification engineers whose role is to evaluate what code is accepted, what it affects, why we should believe it works, and what remains unresolved.

Software engineering is evolving into factory engineering!: An effective AI software factory requires pairing agents with a structured engineering "harness" rather than granting them unrestricted autonomy. Existing platform tools (CI/CD, security, observability) serve as the underlying machinery, while the harness provides context, control, and accountability.

Dear Coder: Open This If You're Feeling AI FOMO: Developers should not allow the hype of AI to replace genuine engineering knowledge.

DESIGN.md vs. a Design System MCP: Which One Does What: design.md is the strategy and rules, the source of truth, while the design system MCP is the implementation of design.md extended for code implementation, keeping both context files focused and maintained via CI/CD.

Why Architecture Frameworks Matter: Architecture frameworks still matter because they reduce blind spots and provide structure for decision-making without replacing the judgement of the architect.

AI Agents Can Think. Now They Can Pay: Agents are now becoming economic participants, which means they need a way to pay for services without relying on human checkout flows. This has led to the introduction of the Machine Payments Protocol (MPP), a way to make machine-to-machine transactions.

Maintaining quality while agents write the code: To prevent hidden design debt in large pull requests, teams must enforce explicit human ownership, lightweight decision records, and realistic review limits.

Why the Best Software Advice Is the Hardest to Follow: Engineering maturity means moving beyond adherence to rules. It starts with understanding why standards exist, evolves into knowing when and how they apply, and culminates in using accumulated patterns and trade-offs to exercise sound judgment

How to answer "Just use AI and move faster": Managers should act as a bridge between business expectations and engineering reality, using hands-on experience and concrete trade-offs rather than simply shielding teams from pressure or blindly demanding more AI adoption.

The word of the day is Déjà vu,. Déjà vu describes the illusion of having previously experienced something encountered for the first time.

Example in a sentence:

Walking into the unfamiliar lab, she was struck by a sense of déjà vu.

I am making my grand departure into the unknown.

Take care of yourself!

Until the next fortnight, my treasured reader, go forth, and may the odds be ever in your favour! 👏 🤖 ✊ ☠️ 🏹 🪖