TSBT60: Exploration
Thoughts and pieces of internals
My dearest gentle reader, I hope you are doing your best to live as well as you can amid the current events in the world.
I recently wrote a LinkedIn post covering CPython Internals on memory allocators that you should definitely read to understand the inner workings of the language we know and love. You can read it here and share your thoughts, too, on the interesting internal things you are exploring.
Yes, and… Carson’s essay argues that while AI is transforming the tech field, computer programming skills remain essential for managing complexity and problem-solving. AI is a good teaching assistant that can help developers bypass technical blockers, but the danger comes in when developers use it to generate all their code, especially for juniors. He shares that writing code is the best way to develop deep literacy in the code required to read, debug, and architect systems.
One of the tools I use to figure out colours in the development process is Happy Hues, designed by Mackenzie. It helps you see the different contrasts in different colours. It provides different colour palettes that simplify decision-making. I highly recommend it, especially if you enjoy the more pragmatic colours.
I follow an architecture principle I call The Law of Collective Amnesia: this article argues that systems eventually drift away from the original design due to factors such as team changes and people forgetting the architectural intent behind earlier decisions. While documentation and institutional memory preserve intent, the author argues that systems should be designed so that the correct architectural path is the easiest and safest to follow. What this looks like is enforcing clear boundaries and structuring the systems so developers naturally extend them in the intended way rather than accidentally breaking them. The emphasis is on the architecture to automatically guide and constrain future development, preventing the system's decay caused by collective amnesia.
How microservice architectures have shaped the usage of database technologies: microservices have changed how organisations use databases. Instead of choosing between SQL and NoSQL, companies now run multiple databases side by side, depending on workloads. This has led to more fragmented data, prompting the adoption of data integration layers and event-driven architectures to synchronise data across services.
SKILLs MD for Analytics: How We Made Snowflake Intelligence Agents Reliable for Production: The team made Snowflake Intelligence AI agents reliable enough for production analytics by encoding analytical workflows as “SKILLs”, structured, version-controlled procedures that define inputs, logic, validation rules, and guardrails. Instead of relying on long prompts or multiple specialized agents, they built one agent that selects and executes the appropriate SKILL, ensuring deterministic and auditable results. These SKILLs are developed with analysts, stored in Git, automatically deployed to Snowflake, and executed through a system that separates skill discovery (search) from skill execution (structured loading). The main gist is that turning analytical knowledge into explicit, versioned procedures prevents hallucinations and makes AI agents dependable for real business analytics.
11 Things I learned after using AI Agents full-time: productivity improves when developers treat AI as a structured collaborator. Some ways to use AI as a collaborator include planning tasks before execution, maintaining persistent context rather than isolated prompts, using reusable “skills” for common workflows, and breaking work into small, focused tasks rather than a single huge prompt. The author also stresses the need for human oversight, such as carefully reviewing code, setting rules and guardrails, learning the tools deeply, and keeping tests and CI in place, because AI can generate code quickly but also accumulate technical debt just as fast.
Is it over for metrics?: This article argues that metrics may be becoming less important because modern systems can store and query raw event data cheaply, allowing teams to compute metrics on demand instead of pre-defining them.
Software Acceleration and Desynchronization: The author argues
that speeding up parts of the software development process can unintentionally disrupt coordination among the many feedback loops involved in building and operating systems. Activities such as coding, reviews, operations, platform work, and learning from production run in interconnected cycles that periodically synchronise. When one loop, such as code generation with AI, accelerates faster than the others, these loops drift apart, a phenomenon the author calls desynchronization. This makes teams appear faster in the short term but increases surprises, duplicated work, and operational risk because important feedback and shared understanding arrive too late. Some “slow” steps, like reviews or operational feedback, exist to keep the system synchronised, and removing them may speed up local work while harming the overall system.
I enjoy essays of various forms and topics; feel free to share some of the essays you have found interesting.
The word of the day is libertine. Libertine refers to a person who is unrestrained by convention or morality.
Example in a sentence:
You can't expect a libertine like Shawty to enjoy a structured, nine-to-five office job; she thrives on chaos and spontaneity.
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! 👏 🤖 ✊ ☠️ 🏹 🪖
Do you enjoy the latest issues of my newsletter? Buy me a piping cup of hot chocolate today!
