Thanks @anthropic.com!

Welcome to a mid-week helping of the August dotInsights newsletter! Highlights this month: 🚀 A gentle introduction to Git worktrees: @humanwhocodes.com 🧠 C# closed class hierarchies: @andrewlock.bsky.social 🛠️ Async/await with @louella.dev Read & subscribe here: blog.jetbrains.com/dotnet/2026/...

dotInsights | August 2026 - The JetBrains Blog

Did you know? Ever wondered why C# uses the semicolon to terminate a statement? Obviously it comes from the C/C++ heritage it shares with other languages like Java, but where did C get it from? That w

blog.jetbrains.com

Grok 4.5 has so far been underwhelming for me. It really struggled with the task I gave it, introduced repeated syntax errors, and couldn't figure out how to get to the final state. GPT 5.6 Luna sailed through the same task without breaking a sweat.

GPT 5.6 Luna's weak spot in my workflow is SQL. It has a lot of trouble figuring out permissions and needs multiple turns to get it right. Will use another model for this from now on.

npm has finally taken steps to protect the ecosystem from malicious packages that don't push more work onto maintainers. I'll be discussing in my newsletter on Tuesday. There's still time to sign up to get it in your inbox: https://newsletter.human...

Man, trying to get WatermelonDB to compile with Expo 56 is brutal. All of the documentation is out of date.

Ran out of space on my Windows laptop. While I'm waiting for my Framework 13 Pro, I loaded Ubuntu onto an external drive and it's been a joy. Tip: Don't install dev tools using snap, stick with apt.

A friend's company is looking for a senior developer to work on their Django/Svelte application. Full-time remote on a small team where you can make an impact. https://hello.fundamatic...

After seeing the first reviews of the Framework 13 Pro, I'm getting excited to receive mine. I'm in batch 8 and really looking forward to it.

tfw when your app’s test build has been approved.

The latest version of VS Code crashes my WSL instance roughly once every 15 minutes. Pretty tough getting anything done since the update.

LLMs are good at making things work when they're broken. I've yet to find a model that will step back and look for ways to avoid similar problems in the future.

Last week of the month and my OpenCode budget is blown. Copilot is at 92% of usage. The rest of the week is looking less productive.

I asked GPT 5.4-mini and GPT 5.6 Luna to do the same task. Luna (medium effort) accomplished the same result in 50% fewer tokens than mini.

Great intro to git worktrees from @humanwhocodes.com : humanwhocodes.com/blog/2026/07... The GitHub Copilot app uses them by default, and it reallly helps to understand them, since a worktree will *not* have files from the main checkout like a gitignored .env.

A gentle introduction to Git worktrees - Human Who Codes

Git worktrees let you check out multiple branches into separate directories simultaneously, making parallel local development straightforward whether you're working solo or alongside AI coding agents.

humanwhocodes.com

I built Bredbox on Supabase and I have some regrets. Without upgrading the npm package, they changed the Docker images and my whole local setup broke. Local setups should remain working as long as I don't change the package!

My latest process: 1. GPT-5.6 Terra writes the PRD 2. GPT-5.6 Terra writes the tech spec 3. GLM 5.2 creates the implementation plan 4. DeepSeek V4 Pro follows the plan MAI-Code-1-Flash and GPT-5.6 Luna for incremental changes.

npm staged publishing is useful, but the workflow is high-friction. For one or two packages it's fine; for dozens, the overhead is brutal. I'd rather require 2FA on every publish: approve a text prompt and continue. No extra website logins.

GLM 5.2 is really good at drafting plans and fairly good at drafting tech specs. However, it's really inefficient add editing tech specs. $2-3 per revision. It's thorough, but other models can make revisions more cheaply.

More than 200 beta users have saved over 10,000 links to Bredbox. Request your invite here: bredbox.app

GLM 5.2 put together an impressive implementation plan for a tech spec. Not only did it break everything down, it also compared against the codebase to find inconsistencies and asked whether I wanted to make adjustments based on that.

Man, Expo is so janky. It looks like local iOS builds are completely broken on Mac Tahoe. https://github.com/expo/...

I don’t code on weekends. I haven’t in over ten years. Try it. Let your brain relax a bit.

This week's Bredbox updates are now live! 📥 Saves is now called Inbox​ to make its purpose clear 🗑️ Deleted links now end up in Trash for 30 days so you can restore them easily ⚡ Fixed a bug that removed at least 2s from each metadata fetch 📱 Test build of the Android app awaiting publishing

Don't sleep on MAI-Code-1-Flash for making small changes. It's incredibly token efficient and now my go-to as I'm making small, iterative changes with Copilot.

GPT-5.6 Luna is blowing me away. What it can get done in fewer than 25 credits is game-changing.

One of the things I find is that all models are overly defensive when it comes to input validation, often creating a bunch of unnecessary 'if' statements. This code wants to validate ?permanent=true in a query string. Anything other than "true" should be false. AI generated code and my edits below.

There's a special place in hell for companies that make me spend 10 minutes navigating a phone tree only to tell me to use the website to complete this action.

Really liking GPT-5.6 Terra. It's much less sycophantic than 5.5 and 5.4. I asked it to make a change that didn't make sense (I had forgot a detail) and it asked if I was sure because of those details. That would have been an annoying mistake otherwise.

In case you missed it, my introduction to git worktrees is now live.

Nicholas C. Zakas@humanwhocodes.com · 3w ago

You need to understand git worktrees in 2026. All of the writeups I found confused me. They were either too high level or always focused on AI. So I wrote my own gentle introduction to git worktrees: https://humanwhocodes.co...