Data Loves Structure • The Applied Go Weekly Newsletter 2026-11-22

Your weekly source of Go news, tips, and projects

Data Loves Structure
Hi ,
Data structures! The foundation stone, cornerstone, and capstone of any serious application. They're so central to application architecture that it's always worth learning more about classic and new ones and the systems built upon them.
A recently accepted proposal underlines the importance of data structures: Coming in Go 1.28, types don't have to be typed out when writing a composite literal; the Go compiler infers the type from the declaration. (Featured article #3.)
Data structures also play a vital part in Entity Component System (ECS), a data modeling approach for games that considers every game a combination of, well, entities, components, and systems. (Featured article #2.)
And sometimes, aligning a data structure such as a loop header with the beginning of a cache line can speed up a program. (Featured article #1.)
What's the next data structure you want to learn about?
–Christoph
Featured articles
Inside PGO | Internals for Interns
Where does my program spend all of its time? Profile-guided optimization (PGO) reveals the time-consuming parts and asks the compiler to optimize these parts better.
Making Games in Go: Entity Component System (ECS)
"Focus on the process instead of the outcome. Make a game your way. It’ll make you want to scream “LOOK WHAT I HAVE CREATED!”" – Miłosz Smółka
Accepted proposal: omit the type in composite literals
The Go team isn't exactly known for rushing proposals through the approval process. This is one factor of Go's legendary stability and backward compatibility. So it can happen that a proposal from 2015 gets approved eleven years later.
Spotlight: Your Craft, Your Rules: What to (Not) Hand Over to AI
In the previous article, I took the pieces of my LLM-damaged brain together to analyze the detrimental effects of delegating brain work to the bots. Brain rot is an insidious malaise, and the remedies require effort and a good dose of self-control. It's almost always easier if you have some patterns to stick with. Which pattern is best depends on many factors; below I list five of them to try out.
Five patterns that set clear boundaries
There are many ways to interact with LLMs on projects; these patterns are examples of where to draw the line between human work and bot work.
More importantly, they protect your authorship. You can still say, "I worked on this project."
Spec-first
The spec-first pattern assigns all work on the specification to the human and all implementation work to the LLM. This one is perhaps the most rigid divide. All your analytical and creative skills go into building the grand picture and painting some rough details. You will let go your coding skills this way, so be sure to accumulate new skills on the specs side.
Test-first
Here, you go one step further. Set up the project layout, then build the package APIs as function stubs (only the signature, no function body). Then write behavioral tests for each function. The bot's task is to complete the functions so the tests turn green. (TDD practitioners, I know you yawn at me. You know this workflow already.)
AI, the junior
If you aren't a fan of strict borders, try treating the bot as a junior partner at pair programming. You take the thinking part; if the bot was a human, you would stand behind its chair breathing down its neck while thinking aloud about the design of a data structure or the implementation of a function. The bot does all the typing.
AI, the critic
There are situations in which you don't want to hand any programming over to the bot. Let the old clanker be your critic instead! Throw your steaming fresh code before its feet and ask it to take it apart. Let it find inefficiencies, bugs, untreated edge cases, and security holes.
AI, the explainer
Writing code is fun, wading through an ocean of foreign code isn't. To get a foot in the door when analyzing other people's (or bots') codebases, ask the bot to explain it: Let it analyze the technical architecture, show you the layers, or give you a linear walkthrough of a particular part of the app. Bots are happy to explain things (or rather, they simulate being happy), and so they can serve as a second pair of eyeballs if your teammates are otherwise engaged as usual.
Avoid these three anti-patterns
Uh, what a clickbait-y heading. In my defense, it's quite a way into the article, so it surely won't trigger any extra clicks.
Anyway, there are a few things that some YOLO-minded vibe coders swear by but that can actually become quite counter-productive, at least in the long run.
No agent-driven planning
Don't let agents plan out all of your project. LLMs lack intention; all they can do is calculate the most probable answer to an input. This hasn't changed from GPT-3 to Mythos 5.2. Only the sophistication in doing this calculation has greatly increased. Without intention, LLMs cannot pursue higher goals; they only can go after technical goals, if they have a definition of done at hand. You might get something that works, but maybe not what you intended.
No merges on autopilot
"We're drowning in PRs since we've let the bot chomp through the tickets! Let's have another bot review the PRs and merge them!" Bad idea. This is for the adventurous who have no problem fixing messy code that did not go through any meaningful quality gate.
No autocomplete while thinking
LLM autocomplete is super okay! Or are you really looking forward hacking that boilerplate code into the keyboard for the umpteenth time? But when you are working on a difficult algorithm or a design of a complex data structure, a delicate SQL statement or code whose correct implementation needs a lot of background knowledge, turn autocomplete off to keep your brain sharp and the bot from presenting average suggestions that mess with your train of thoughts.
Don't become the reverse centaur
The "human plus AI" combo is often described as a centaur: a human's head and torso on top of a horse's body. The human contributes intelligence, the horse contributes power. Cory Doctorow observed the opposite happening in many companies: AI is the software team's brain, humans are degraded to mere click workers, rubber-stamping every PR because they lack the time reading through monster PRs that come in at a neck-breaking rate. He coined the term "reverse centaur" for this constellation.
Imagine how this reverse centaur looks like. Does the image now stick in your brain? Good. Keep it there. It'll remind you of the importance of remaining in control.
Your craft, your rules.
Quote of the Week: Do we really live in the future?
I’m reminded of that meme where Tyler the Creator says “how the fuck is cyberbullying real, just walk away from the screen.” Do we really live in the future if it doesn’t exist outside of our computer?
More articles, videos, talks
One Go App, Served Five Ways · JustinFX.com
The new way of prototyping: "Hey, clanker, build me five versions of an an app with five GUI libraries!"
What the GitHub Secure Open Source Fund changed in modernc.org/sqlite
modernc.org/sqllite is one of the alumni of the GitHub Secure Open Source Fund. The maintainers took part of a three-week security training. Here, they share how users of the library benefit from this training. (Plus some tips for maintainers of similar projects.)
Incident pipelines with Spooler
Pagedog receives incident pages via webhooks and runs LLM investigations on them. In this article, Sergey Kamardin uses Pagedog to demonstrate how to build durable pipelines and how an atomic operation helps connecting the queues.
Probabilistic data structures in Go, part 2
Anton Zhiyanov has more probabilistic data structures for you, probably.
Projects
Tools and applications
OpenTodoApp/OpenTodo: Your tasks, your server. An open-source, local-first task manager you can host yourself. Everything you expect from a commercial task app, without the subscription or the lock-in.
A task manager service without subscription (if you have a server to host it).
Feasible-Analytics/app.feasible.lol: Privacy-first web analytics with no cookies. One Go binary, one SQLite database per account. Hosted or self-hosted.
Feasible-Analytics has everything except cookies: A dashboard, goals and funnels, search term analysis, reports, alerts, and more.
Updates
Shirei v0.9 | Judi Systems
Shirei is a cross platform GUI framework. v0.9 adds native auxiliary windows and overlays, a document editor, and furigana readings.
Completely unrelated to Go
Do Repeat Yourself · Arcentry
The DRY principle works quite well within a codebase (but keep the Go proverb in mind: A little copying is better than s little dependency). On the teams level, DRY can fail miserably.
The Internet’s Doomsday Prepping Spiral - by Saanya Ojha
Dark forces cause all the world's trouble! Better order your personal prepper kit from Amazon before the next coordinated AI attack leaves us all sitting in the dark...
