Hello friends!
Welcome to this week’s Sloth Bytes. I hope you had a great week.
Speak naturally. Send without fixing.
Wispr Flow turns your voice into clean, professional text you can send the moment you stop talking. Not rough transcription you have to clean up. Actual polished text — ready for email, Slack, or any app.
Speak the way you think. Go on tangents. Change your mind mid-sentence. Flow strips the filler, fixes the grammar, and gives you text that reads like you spent five minutes writing it.
89% of messages sent with zero edits. Millions of professionals use Flow daily, including teams at OpenAI, Vercel, and Clay. Works on Mac, Windows, and iPhone.
Introducing The First Agentic CRM
Get revenue agents, workflows, and automations across every stage of your motion. Access customer data in real time through Attio's web app, MCP, API, and SDK.
Then Ask Attio anything about your business and get instant answers.
It's the CRM that runs the work behind every win.
Bun 1.4 is out!
Bun shipped 1.4 and it’s the first release with the Rust rewrite. Claude Code has been running on it for months.
Biggest Node jump since 1.0. +1,517 tests from Node's own suite and over 2,900 issues fixed.
Idle CPU is down 5x. Memory is down up to 35%.
Linux
hello.jsstarts in 5.1 ms (was 10.9). Windows is 15.5 ms (was 39.0).node:http,node:fs, and friends pass 97% of Node's tests.Playwright and vitest finally run.
bun --bun next buildworks on Next.js 16.3.New goodies to mess around with:
Bun.Image,Bun.WebView,Bun.markdown,Bun.cron(),bun run --parallelandbun audit fix.
Another win for Rust… Great. Rust users aren’t gonna shut up about this for a while.
Apple put M6 in the Mac mini
Apple unveiled a new Mac mini after 2 years, with M6 and M5 Pro chips. Pre-order started August 25 and they'll ship on September 22.
M6 Mac mini starts at $899 (US), $799 for education. M5 Pro starts at $1,699 (US), $1,599 education.
Apple's claims vs the M4 Mac mini: up to 4x faster AI, 2x faster storage and graphics, 40% faster CPU.
M6: 12-core CPU, 12-core GPU with Neural Accelerators in each GPU core (first time on Mac mini), Dual 16-core Neural Engine, up to 170GB/s memory, 16GB standard unified memory up to 32GB.
Pitched for on-device AI and always-on agentic workflows. LLM prompt processing in LM Studio: Apple says up to 4.8x faster than M4, 13.5x vs M1.
Wi-Fi 7, Bluetooth 6, 2.5Gb Ethernet (10Gb option). M6 has Thunderbolt 4. M5 Pro has Thunderbolt 5.
macOS 27 Golden Gate comes with it (public beta now, free update this fall).
Are you gonna upgrade your local AI assistant with this? I know I won’t.
Debian is voting on whether AI can be used
Debian is voting on whether AI contribution should be allowed and voting ends tonight.
Why this is happening: LLM-written packages and docs started showing up. The ban camp says copyright is unclear, the quality is sloppy, new contributors dump slop on reviewers, and the scrapers already knocked Debian's own sites over. Other people already use the tools and don't want a ban they can't enforce.
The options: There are 8 options, but for the most part they’re in this area:
Hard ban: No LLM-written contributions. That one needs a 3:1 majority because it changes the Social Contract.
Allow it with rules: disclose when you use it, but stay accountable.
"Debian is created by humans": The model's output can't be the contribution, researching with one is fine.
Or a climate statement: avoid LLMs because the energy cost is a deal-breaker, but no ban/punishment.
Node.js picked the created by humans option: Use the tools to research and understand the codebase, but never to write code.
AWS is buying the company behind DuckDB
AWS is buying DuckLabs! Man I love that name.
They’re Amsterdam company behind DuckDB and this will take affect early September.
Current DuckLabs project s will stays open source.
DuckDB, DuckLake, Quack, and the extensions stay MIT.
The nonprofit DuckDB Foundation keeps the IP.
They’ll get to stay in Amsterdam and keep pointing the technical direction.
They get more than a million downloads a day, from a bootstrapped team of 30+.
They picked AWS because they didn't want to become a sales org and stall the project.
AWS is already running it inside Amazon Quick: over 2.5 billion queries since October 2025, and they claim a 30% drop in average query latency.
Maybe AWS will come out with a service called AWS Duck. That’d be cool.
Shopify's CEO vs Claude file
I’m so glad someone is talking about this.
Tobi Lütke (Shopify CEO btw) is thinking about banning Claude Code at Shopify.
And not because the agent is bad, but because Claude Code doesn’t support an agents.md file. WHICH IS VALID. I HATE THIS TOO. They’re the ONLY AI tool that doesn’t support this.
AGENTS.md is the format Codex, Cursor, Copilot, and a bunch of other tools already read.
Claude Code only accepts a CLAUDE.md, but you can link the agents file within it.
The reason for this complaint is a split-brain problem when developers use different tools. He describes it as a “lobotomy.” Beautiful way to say it.
Shopify has to do an automation to handle this, but Tobi calls it a stupid complexity tax. Which is 100% reasonable.
Anthropic's workarounds: import AGENTS.md from CLAUDE.md, or use symlinks, but WHY NOT JUST SUPPORT IT?
Thariq from Anthropic responded saying they’re making Claude Code more “hackable.” Which means Claude Code will soon SUPPORT THIS. Can’t wait.
Is Front End Education dying?
NOT design btw. Just front-end.
Nolan Lawson thinks front-end education is standing in a crater. A lot of front-end educators he admires are leaving or pivoting to AI.
He thinks the future of front-end education is not looking good and even front-end development itself will be interesting since AI is reaching that “good enough” stage.
“…even as someone who’s written about browser performance for years and who literally worked on a browser performance team, I would probably just chuck a Chrome trace at Claude Code and have it suggest improvements. In fact I’ve done this very thing in my day job and gotten some good results.”
Handing a React component to an agent is lower risk than a database migration, so a lot of AI coders will yolo the frontend unsupervised.
Education will shift from front-end to other areas:
Teaching AI agents the big picture (system design and which tools to actually use)
Make sites that work for agents
Fixing bad patterns from vibe coded apps
Now this doesn’t mean front-end education or front-end itself is dead. It just means maybe what is being taught will be completely different in the future.
I feel like it’s the same as other areas of programming where it’s no longer about the syntax, but the patterns and architecture.
What do you all think?
How Uber built an AI software factory - Uber says more than 70% of their PRs now come from agents so they built a system that doesn't melt their CI process.
When str.lower() is a security vulnerability - I know the title sounds crazy, but it really shows how many edge cases you have to think about.
How to report a bug so it actually gets fixed - A really useful guide to creating effective bug reports.
MicroLighter - A ~2KB highlighter that paints tokens with CSS Custom Highlights instead of wrapping every token in a span. Great way to add a highlighter effect to your projects.
Formisch 1.0 - A cool framework-agnostic form tool. You build your forms with a schema which gives you type safety. Also great performance!
TermDOM - Really interesting project. You create terminal apps like you’re writing HTML, CSS, and the DOM.
hk - Git hook manager that runs linters in parallel without letting overlapping fixes race. Didn’t even know this was a problem. Maybe you’ll need it.
That’s all from me!
Have a great week, be safe, make good choices, and have fun coding.
If I made a mistake or you have any questions, feel free to comment below or reply to the email!
See you all next week.
What'd you think of today's email?
Want to advertise in Sloth Bytes?
If your company is interested in reaching an audience of developers and programming enthusiasts, you may want to advertise with us here.
