Hello friends!
It’s me again with another Sloth Bytes. I hope you’re doing amazing.
For the people with low attention spans:
Jev launched - a new AI model that skips the chat and hands your code a probability instead. Weird, but kind of brilliant.
Homebrew 7.0 - real Linux sandboxing, a built-in vuln scanner, and finally a native macOS app.
Copilot's Rust rewrite - ~800k lines, one developer, mostly written by AI.
A fruit fly's brain is now playing Doom - yes, really, and it only took a week.
200+ Proven Ways to Make Money With AI in 2026
The next wave of millionaires will be people who figured out how to make AI work for them.
The window to get ahead is still open. But not for long.
Here are 200+ proven ways to make money with AI in 2026.
Sign up for Superhuman AI, the free daily newsletter read by 1M+ professionals, and get instant access to all 200+ ways to profit from AI this year.
The Future of AI in Marketing. Your Shortcut to Smarter, Faster Marketing.
This guide distills 10 AI strategies from industry leaders that are transforming marketing.
Learn how HubSpot's engineering team achieved 15-20% productivity gains with AI
Learn how AI-driven emails achieved 94% higher conversion rates
Discover 7 ways to enhance your marketing strategy with AI.
New type of AI model released called Jev
This is NOT just another LLM, it’s actually a whole new type of AI model.
Most models want to write you an essay, even when you ask it a simple yes or no question. They’re not the best when it comes to fast, structured decisions.
Jev is an attempt to fix this problem by rethinking the way these AI models are made.
The actual difference: a normal LLM generates its answer one token at a time, optimized to sound fluent. Ask it for a probability and it might just yap an entire essay instead of the number you were expecting. Jev skips text generation entirely and computes every output in parallel in a single pass, so what comes back is a probability or a category.
Why it's trained differently: normal LLMs are shaped by RLHF or RLVR, reinforcement learning that rewards answers humans like or that pass a verifiable check. Jev is trained with what TypeSafe calls RLCD, reinforcement learning for calibrated decisions, so the reward is how honest the confidence number actually is, not how good the sentence sounds.
Speed: Not gonna lie, Jev is BLAZINGLY FAST. Apparently it answers in 70 to 500 milliseconds.
Keep in mind, current frontier LLM models take 3 to 329 seconds
Cost of this model: This is the insane part you won't believe.
Input tokens: $0.042 / MTok ($42 per billion tokens), versus $0.20 to $10 per million for frontier models.
Output tokens: FREE (too cheap to meter) - literally what they say. TypeSafe admits it can't yet prove that stays sustainable, so don't bet your margins on it staying this cheap forever.
Why it matters: if your product calls a frontier LLM to classify a support ticket, route a lead, or pull fields out of an email, that's exactly the job Jev is built to replace, and at these prices you can afford to run the check on every request instead of doing workarounds to save money.
Potential use cases: Their doc page sums up the insane potential.
Homebrew 7.0 is here and it’s big
Homebrew 7.0 shipped and it's a bigger release than the version number suggests.
Speed: installs and upgrades now overlap downloading and preparing packages instead of doing them one after another.
brew cleanupalso stops re-scanning the cache from scratch every time.
Security: Linux sandboxing moves from Bubblewrap to Landlock (no extra deps, no more Docker permission headaches)
plus a new built-in
brew vulnsscanner that checks your formulae against a fresh Homebrew advisory database on OSV.dev
New app: BrewUI, Homebrew's first real native macOS app, is fully released.
GitHub rewrote the Copilot runtime in Rust
It only took a few months, but one developer pulled this off thanks to AI.
The blog post is definitely worth the full read. It's very detailed even for an engineering blog post.
How much code: 800,000+ lines of production Rust, shipped across 128 PRs.
The reason: GitHub's writeup lays out exactly what they wanted instead of the old Node runtime:
A runtime split cleanly from the TUI, so it's just a library other tools and services can layer on top of
Minimal dependencies, minimal overhead
Embeddable in-process instead of forced out-of-process
Real performance, scalability, and reliability characteristics
Clean interop across all six Copilot SDK languages (C#, TypeScript, Python, Rust, Go, Java), each through its own FFI
A toolchain with a stronger security posture: less supply chain risk, more room for correct-by-construction code
The honest bits are my favorite:
Agents spent way more time reading than writing
Prompt-cache hits were critical for cost
"if it compiles, it is correct" is NOT true.
Google mapped a fruit fly's whole nervous system and developers immediately made it play Doom
This is just a fun one I wanted to share because of what programmers did with it.
Google Research and HHMI Janelia mapped every neuron in a fruit fly's brain and published the whole thing, free to download.
Since the internet is well… the internet. They used this data to do dumb things.
Yeah. I think we’re cooked if we ever fully map a human brain.
Five software engineering roles for working with AI - Breaks down five ways to work with agents, from barely touching the code to reviewing every line. Worth figuring out which one you actually are.
The Cost of Abstraction for Humans and AI Agents - turns out over-engineered codebases don't just annoy your teammates, they burn through your AI budget too. About 30% more, by this post's estimate.
I made my website charge AI agents a penny per page. Then I watched Claude pay it. - genuinely one of the funnier experiments I've seen this year, and it actually worked.
We are all Product Engineers now - A take on what's left for devs to do once agents handle writing and reviewing all the code.
Analytics on Live Data Without Leaving Postgres
When analytics on Postgres slows down, most teams add a second database. Then they manage pipelines, sync lag, and drift forever. TimescaleDB extends Postgres instead. Analytics run on live data, in the database you already have. No pipeline. No migration. No new query language.
Worktrunk - git worktrees without the ceremony, handy if you're running a few agents in parallel and don't want to type the branch name three times.
Google's modern web guidance skills - Google shared some skills for web development best practices like UX and performance.
neat-annotations - hand-drawn CSS arrows and notes, zero JS, one stylesheet. Good for docs that need fun labels.
pdfcn - Some useful customizable PDF components for React.
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.
