In partnership with
Hello friends!
Welcome to another edition of Sloth Bytes. I hope you had a great week.
Your prompts are leaving out 80% of what you're thinking.
When you type a prompt, you summarize. When you speak one, you explain. Wispr Flow captures your full reasoning — constraints, edge cases, examples, tone — and turns it into clean, structured text you paste into ChatGPT, Claude, or any AI tool. The difference shows up immediately. More context in, fewer follow-ups out.
89% of messages sent with zero edits. Used by teams at OpenAI, Vercel, and Clay. Try Wispr Flow free — works on Mac, Windows, and iPhone.
Microsoft Build 2026 happened and there’s potential
Two and a half hours of Satya Nadella saying "agentic." Here's what's actually useful.
GitHub Copilot gets a standalone desktop app (technical preview)
One dashboard for all active agent sessions, PRs, issues, and background automations
Each session runs in its own isolated Git worktree so you can runparallel agents with no conflicts.
Agent Merge monitors CI, tracks reviewers, and carries PRs to merge on your behalf
Canvases keep agent decisions visible instead of buried in chat
Local and cloud sandboxes so agents iterate without touching production
MAI-Code-1-Flash is live in the VS Code Copilot model picker today
A reasoning model (MAI-Thinking-1) is in private preview if you want to request access
Windows Developer Configurations: one WinGet command sets up VS Code, Copilot, WSL, and PowerShell 7 on any blank Windows 11 machine
Coreutils for Windows: Linux-like command line utilities (ls, cp, grep) running natively on Windows without WSL
WSL Containers: run Linux containers with familiar CLI/API, no separate VM. Coming to public preview soon
Intelligent Terminal (experimental): context-aware agent built into your terminal that can read errors, debug, and run multi-step tasks without breaking your flow
MXC (Microsoft Execution Containers) (early preview): declare exactly what an agent can access at the OS level. Isolation enforced at runtime, not vibes
Angular v22 is out!
New version of Angular dropped (if you use it) and the experimental APIs are over:
Three big APIs hit stable: Signal Forms, Angular Aria, and the async reactivity APIs are now stable.
OnPush is now the default change detection in components
ChangeDetectionStrategy.Defaultis also renamed toEager, which is more honest about what it does.running
ng updatewill handle the migration
New
@Servicedecorator: Replacement for@Injectable({ providedIn: 'root' })and now the default output ofng generate service.injectAsyncfor lazy-loaded services: Routes and components could already lazy-load, but now dependencies can too. Useful for heavy dependencies you only need after a specific user action.New Angular MCP tools: New features were added to the MCP where Agents can interact directly with your dev server:
devserver.start,devserver.wait_for_build, anddevserver.stopAngular Agent Skills: This give AI assistants up-to-date knowledge of Signal Forms and Angular Aria that their training data probably doesn't cover yet.
Lot’s of other stuff, but these are probably the biggest.
Cloudflare Just Acquired Vite
Cloudflare acquired VoidZero. They’re the company behind Vite, Vitest, Rolldown, Oxc, and Vite+, so basically every modern JavaScript build tool.
Why it happened: Monetizing open-source tooling proved too hard. Mixed licensing didn't feel right. So they open-sourced everything under MIT and started building Void, a Vite-native deployment platform on Cloudflare. That experiment made the acquisition obvious
What stays the same: Vite, Vitest, Rolldown, Oxc, and Vite+ remain MIT-licensed. The same team continues leading all of it
The money commitment: Cloudflare is putting $1 million into an independent Vite ecosystem fund for open source maintainers.
Benefits for Cloudflare: One-click deploys from Vite straight to Cloudflare's global network and better tooling for AI agents as Cloudflare positions itself as the cloud for agents.
Looks like AI is getting too expensive for companies
The "Tokenmaxxing" era might be ending, yippee.
Big companies spent the last year treating AI token usage like a flex:
Meta built an internal leaderboard called "Claudeonomics" ranking its top 250 token burners across 85,000 employees, handing out badges like "Token Legend" and "Cache Wizard"
Uber put engineers on a scoreboard to see which teams used AI the most
The logic was basically more tokens = more productive = employees make company rich.
Guess that didn't work. (What a surprise...)
Meta took down its tokenmaxxing leaderboard two days after news of it broke publicly
Microsoft canceled Claude Code subscriptions across several key product divisions
Uber burned through its entire 2026 AI budget by April
Walmart ended unlimited token access for its internal AI assistant Code Puppy, moving to fixed per-employee allocations
And one unnamed company reportedly spent $500 million in a single month after simply forgetting to set a usage cap
It's not that the tools don't work. It's that nobody can prove what they're producing. Uber's COO said the company has yet to draw a clear line between its Claude Code spending and delivery of new customer-facing features.
There's a 4,500x pricing spread between the cheapest and most expensive AI models today. Most companies defaulted everyone to the newest model for everything, so an easy optimization would be a smart router that would decide when to use an expensive model.
Why it matters for you: This is mostly a big-company problem right now, but it's easily gonna spread to all companies. If you're using Claude Code, Cursor, or any agentic tool, the new skill isn't how much AI you can spam. It's how smart and effective you can use it.
How we reduced core unit boot time from hours to minutes - Cloudflare's team digs into how a routine firmware update caused their core servers to take four hours to reboot instead of minutes.
Coding Agent Horror Stories: The rm -rf ~/ Incident - Docker documents the most cursed category of AI coding agent failure: an agent deleting an entire home directory in a single command.
How To Save 90% of Claude Code Token Usage - Given the developer trend above, this one is extremely useful. Four free strategies to save on tokens.
Modern Engineering Values - Christoph Nakazawa (Meta/Jest) writes about what engineering principles actually matter in 2026. "It's funny when people call out how much slop agents are producing. They must have never seen the slop engineers produce at big tech." Refreshingly honest.
favicon-cheat-sheet - The only reference you'll ever need for favicon sizes and types. Obsessively comprehensive.
clean-code-typescript - Clean Code concepts translated specifically for TypeScript. Rewritten for the language you actually use at work.
ytDownloader - A clean desktop app for downloading videos and audio from hundreds of sites. GUI, cross-platform, no command-line required. For those of us who just want a button.
SeaORM - A powerful async-first relational ORM for Rust. If you're building anything in Rust that talks to a database, this is the go-to.
Question for you this week: What's something you've always wanted to build but haven't started yet? Could be stupid, could be genius, doesn't matter.
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.
