Full Stack Focus
Posts
🤖 AI now writes 80% of Anthropic's code

🤖 AI now writes 80% of Anthropic's code

AI is rewriting React's compiler, a designer ditches Figma for Claude, and why every byte still matters.

June 08, 2026

In partnership with

Hey there.

Welcome to Full Stack Focus. 👋

Here's everything you need to know this week in the world of full-stack development.

The best prompt engineers aren't typing. They're talking.

Power users figured this out early: speaking a prompt gives you 10x more context in half the time. You include the edge cases, the examples, the tone you want — because talking is fast enough that you don't skip them.

Wispr Flow captures everything you say and turns it into clean, structured text for any AI tool. Speak messy. Get polished input. Paste into ChatGPT, Claude, Cursor, or wherever you work.

89% of messages sent with zero edits. 4x faster than typing. Works system-wide on Mac, Windows, and iPhone.

Start flowing free

———————————————————————————

Front-End

🪝 The Problem With useEffect
useEffect is the hook React developers reach for most, and misuse most. This breakdown walks through why so much effect code is unnecessary, subtly buggy, or quietly fighting the framework, and what you should reach for instead. The examples are clear enough that you'll spot the pattern in your own codebase immediately. If you've ever chained effects just to keep state in sync, this one will sting a little.

🧹 The ESLint Plugin That Deletes Your useEffects for You
The perfect companion to the piece above: a plugin that flags the effects you don't actually need and nudges you toward removing them. Drop it into your project and watch it light up the spots where you're syncing state the hard way. It encodes the official "You Might Not Need an Effect" guidance into rules you can actually enforce. A genuinely satisfying cleanup tool.

🦀 React Compiler Just Got Ported to Rust — and Mostly Written by AI
Meta is porting the React Compiler to Rust, and the early numbers are wild: roughly 3× faster as a Babel plugin and ~10× faster on the core transform, with all 1,725 fixtures passing. The kicker, straight from the PR author: the architecture was human-designed, but the majority of the code was written by AI. It's a rare, concrete look at how a core piece of frontend infrastructure gets rebuilt in 2026. Worth a read even if you never touch the compiler directly.

———————————————————————————

Programming

🃏 Jira IS Turing-Complete (And Here's the Proof)
Engineering folklore says Jira is Turing-complete — this person actually proved it, by building a working Minsky machine entirely out of Jira automation rules. Registers are counts of linked issues, the program counter is an Epic's status, and yes, they ran addition and Fibonacci on a real Atlassian instance. It's equal parts brilliant and cursed. If your Jira automations have ever felt like programming, it's because they literally are.

🧠 Every Byte Matters
Two loops with the same Big-O can still differ by 30× once you understand how the hardware actually works. This deep dive walks through cache lines, struct-of-arrays vs array-of-structs, and the "memory staircase" you hit as your working set spills out of L1, L2, and L3. It's the kind of low-level knowledge that quietly separates fast code from slow code. A sharp reminder that asymptotic analysis isn't the whole story.

💔 "LLMs Are Eroding My Software Engineering Career and I Don't Know What to Do"
A raw, honest essay from a developer watching AI hollow out the parts of the job they actually loved. It's not a doomer rant and it's not hype — it's the uncomfortable question a lot of devs are sitting with quietly right now. The writing is vulnerable in a way most takes on this topic aren't. Worth reading whether you nod along or push back hard.

———————————————————————————

AI

🤖 AI Now Writes Over 80% of Anthropic's Code
Anthropic laid out internal data on what it calls recursive self-improvement, and the numbers are eye-opening: more than 80% of the code merged into its codebase is now authored by Claude, and the typical engineer ships roughly 8× as much code per quarter as in 2024. It's not all hype, the piece is candid about where humans are still the bottleneck (taste, judgment, and reviewing code faster than AI can write it). A sober, numbers-heavy look at AI building AI. Bookmark this one.

🎨 "I Design With Claude More Than Figma Now"
A designer at Jane Street explains how building real, working prototypes with Claude Code quietly replaced their Figma mockups, spec docs, and design proposals. The honest part is the tradeoff: when reviewers get a fully-baked feature instead of a wireframe, who actually owns the design? It's one of the clearest glimpses yet of where design-engineering workflows are heading. A great read no matter which side of the design/dev line you sit on.

🥊 DeepSeek V4 Pro Reportedly Beats GPT-5.5 Pro on Precision
The latest DeepSeek model is posting precision benchmarks that edge out GPT-5.5 Pro — and reigniting the open-weight-vs-frontier-lab debate all over again. If the numbers hold up under independent testing, it's another sign the gap is closing faster than anyone expected. A quick read to stay current on where the model race actually stands.

———————————————————————————