Sponsored by
Hello gorgeous people
Welcome to another edition of Sloth Bytes. Got lazy this week I’m not gonna lie, but we’re back! This has a mix of last week + this week. Lots of version 7 updates.
Scale AI support on AWS, see how July 9
Customer expectations keep rising. Support budgets don't. On July 9, Fin and AWS are hosting a live executive session on how leading enterprises close that gap: scaling AI-powered support while simplifying how they buy it.
You'll see how to resolve an average 76% of conversations with Fin on AWS enterprise-grade infrastructure, procure through AWS Marketplace to put committed cloud spend to work, and turn the Fin and AWS collaboration into lower support costs. Register for the live session to see how.
Your competitor's growth lead already saw the spend spike.
While your team is still in standup, the other growth lead already got the alert. Viktor is an AI employee that lives in Slack. It watches your Meta and TikTok spend overnight, flags the underperformer by 7am, and drafts the new brief before your first meeting.
Apple Products Just Got More Expensive. WHY.
Apple raised prices across its entire Mac and iPad lineup on June 25th. No new specs, no new models. Identical hardware, higher prices. iPhone, Apple Watch, and AirPods were spared for now.
MacBook Neo: $599 to $699
MacBook Air: $1,099 to $1,299
MacBook Pro: $1,699 to $1,999
iPad: $349 to $449
iPad Air: $599 to $749
Apple TV, HomePod, HomePod mini, and Vision Pro all hit too
The root cause is structural: memory makers like Micron have locked in long-term supply agreements with AI hyperscalers at premium prices, crowding out consumer electronics clients.
If you're thinking about upgrading your dev machine, now might be the best time because these prices might never drop.
TypeScript 7.0 RC Is Here and it's Written in Ru- I mean Go.
Yeah, you read that right. TypeScript 7.0 RC is out, and the entire compiler has been rewritten in Go. This has been in the works since early 2025 under the internal codename "Project Corsa," porting the existing TypeScript codebase out of self-hosted TypeScript/JavaScript and into Go. It's not a redesign from scratch.
The type-checking logic is structurally identical to TS 6.0. They just moved it to a faster runtime.
TypeScript 7.0 is often about 10x faster than TypeScript 6.0, thanks to native code speed combined with shared-memory parallelism. Microsoft tested pre-release builds with teams at Bloomberg, Canva, Figma, Google, Notion, Slack, Vercel, and more over the past year. All reported similar speedups.
A few things to know before you install:
This is an RC, not the final stable release. GA is expected within the next few weeks
Install now via
npm install -D typescript@rc. It ships the standardtscbinary, nottsgoRun it side-by-side with TS 6.0 using the new
@typescript/typescript6compatibility package, which exposes atsc6binaryParallelism is configurable:
--checkerscontrols how many type-checker workers run (default: 4),--buildershandles parallel project reference builds--watchmode got a full rebuild, now based on a Go port of Parcel's file watcher for better cross-platform performanceWait for TypeScript 7.1 before migrating custom compiler API integrations. Stable programmatic API isn't landing until then
Legacy configs like
target: es5andmoduleResolution: nodeare now hard errors. Time to audit that old tsconfig
Why it matters: a 10x speedup on the thing every large TS codebase complains about most. You'll feel it immediately in CI and your editor.
Vercel AI SDK 7: Durable Agents, Tool Approvals, and Production-Ready Workflows
AI SDK 7 dropped and it’s organized around five areas: develop, run, integrate, observe, and go beyond text.
The biggest additions:
WorkflowAgent: the headline feature. Standard agents run in memory and die on process crash. WorkflowAgent runs each tool call as a durable step inside Vercel's Workflow SDK, with automatic retries. If a user closes their browser mid-run, the stream keeps going and resumes when they come back. No Redis or custom pub/sub needed
Tool approval system: agents propose actions before executing. Humans confirm. Your database survives
Typed tool context: scopes API keys and config per tool so third-party tools can't read secrets they don't need
Reasoning control + MCP Apps: a standardized reasoning option controls effort across any provider in one line, and MCP servers can now separate model-visible tools from app-only UI rendered in sandboxed iframes
If you're using v6 you can run
npx @ai-sdk/codemod v7for automated migration, or use the migration skill:npx skills add vercel/ai --skill migrate-ai-sdk-v6-to-v7
Astro 7 Drops with a Rust Compiler, Faster Builds, and AI Agent Support.
Astro 7.0 launched and this one is all about speed. All thanks to Rust.
The .astro compiler got a full Rust rewrite, replacing their previous Go-based compiler.
Which of course makes it BLAZINGLY FAST.
Markdown and MDX processing now runs through Sätteri, a new Rust-powered processor built by Astro core team member Erika, replacing the old unified/remark/rehype pipeline that got sluggish on large sites. The whole thing now runs on Vite 8, which ships Rolldown, a Rust-based bundler replacing both esbuild and Rollup and 10-30x faster than Rollup in benchmarks.
Real-world build time improvements came in at 15-61% faster.
The Cloudflare docs site (8,400+ pages) went from 386 seconds to 261 seconds.
The astro.build site dropped from 62 seconds to 24 seconds.
Markdown-heavy sites see the biggest gains since that pipeline moved entirely to native Rust. The queued rendering engine, experimental in Astro 6, is now stable and the default at about 2.4x faster than the old recursive renderer.
Other stuff worth knowing:
Advanced Routing is stable: gives you a
src/fetch.tsentrypoint to compose Astro features as middleware, with full Hono supportRoute caching is stable: CDN cache providers for Netlify, Vercel, and Cloudflare are now baked in
AI agent support: Astro can detect coding agents (Claude Code, Cursor, etc.), run the dev server in the background via
astro dev --background, and output structured JSON logs for machine-readable feedbackThe Sätteri Markdown processor bundles features that previously needed separate remark plugins: GFM tables, footnotes, smart punctuation, heading IDs, math, wikilinks. All built-in by default
Minimum Node.js version is now v22. Check your environment before upgrading
Upgrade via
npx @astrojs/upgrade
Fable is back and Sonnet 5 came out
New Claude, new number. Claude Sonnet 5 is out and it's now the default model on Free and Pro plans, so you're probably already on it whether you noticed or not.
Price: $2 per million input tokens, $10 per million output. Goes up to $3/$15 after August 31st
Model ID:
claude-sonnet-5via the APICoding: Anthropic's most agentic Sonnet yet, close to Opus 4.8 performance for a fraction of the price. Already the default model in Cursor, where it bumped CursorBench from 49% to 57%
Context window: 1M tokens by default now (used to be opt-in), 128k max output
Breaking change: the old manual extended thinking param (
thinking: {'{'}budget_tokens{'}'}) is gone. Throws a 400 error now if you're still passing it. Adaptive thinking runs by default insteadTokenizer gotcha: the same text now eats roughly 30% more tokens than it did on Sonnet 4.6. If you've got hardcoded
max_tokenslimits anywhere, go recheck them before your outputs start getting silently clipped
Oh yeah and Fable's back. It's available on all your favorite coding tools and will only be available on Claude subscriptions until July 7th, then it will cost usage credits. Also, you can only use it for up to 50% of your weekly usage limits.
Who will be the senior engineers of 2035? - AI is absorbing the small tasks that used to train junior devs, while companies cut entry-level hiring. This piece lays out three scenarios for how that plays out and what engineering orgs should do right now.
How to Write an Effective Software Design Document - Most design docs nobody reads because nobody wrote them well. This excerpt from Refactoring English breaks down exactly what separates a useful design doc from one that collects dust. If you've ever written documentation that got ignored, this is the fix.
Stealing is a Skill - A refreshing take on Virgil Abloh's "3% approach" — copy something almost entirely and let your instincts drift it into something original. Works for code, design, writing. If you're stuck on a creative project, this reframe is genuinely useful.
The Low-Tech AI of Elden Ring - Before ML, game devs were pulling off wild AI lifehacks with state machines and behavior trees. This breakdown of how Elden Ring's enemies "think" is a great reminder that game devs are very, very smart.
replacements.fyi - Type in any npm package and it'll tell you if there's a faster, safer, or native alternative. Kind of embarrassing how many packages I was still installing that have been baked into Node for years.
SSR Checker - Paste any URL and instantly see whether a page is truly server-side rendered and how much content is in the initial HTML. Useful for debugging SEO issues, auditing frameworks, or just being nosy about how other sites are built.
lingo.dev - Open-source AI localization that plugs into your CLI, CI/CD, or GitHub Actions and auto-translates your app whenever you push new text. Works with Claude Code and Cursor via MCP. If you've ever needed to support multiple languages, this is a good one.
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.
