[Dev Weekly #112] Andrej Karpathy Joins Anthropic to Lead Frontier AI | The Real Cost of Agentic Engineering | When Rails Breaks at Scale

HELLO EVERYONE!!! It’s May 22nd, 2026, and you are reading the 112th edition of Codeminer42’s tech news report. Let’s check out what the tech world showed us this week!

The Miners’ post of the week 🧑🏻‍💻👩🏽‍💻

My First Two Weeks as an Agentic Engineer

An honest take on agentic engineering, not the hype version. What got me was how unflinching it is: Claude Code built 600 lines of working code they deleted the same day, hit session limits mid-flow, and struggled with Google Stitch’s design output. But here’s what matters: two shipped projects (Rails 8 app with MIDI integration, Wear OS interval trainer), 288 tests, and a painful lesson about starting a CLAUDE.md file on day one, not day five. The Chrome MCP browser connection detail alone changes how you’d actually work this way. Worth reading if you’re curious whether agentic engineering is real productivity or marketing, because this person shows both the wins and the friction honestly.


Anthropic: Andrej Karpathy Joins To Focus On Frontier LLM R&D — by Amit Chowdhry

Andrej Karpathy, former Tesla AI director and OpenAI founding member, has joined Anthropic to focus on frontier LLM research. He’s stepping back from his education initiatives to dive into what he sees as a formative period for the industry, signaling how seriously top AI researchers view the current competitive landscape in frontier model development.

Staged publishing for npm packages

npm now supports staged publishing, letting you submit package versions for review before they go live on the registry. A maintainer must explicitly approve staged packages with 2FA, adding a safety gate for catching issues before public release. If you manage critical packages or want extra confidence before shipping, this workflow gives you that breathing room.

Mistral AI Acquires Emmi AI to Create the Leading AI Stack for Industrial Engineering

Mistral AI acquired Emmi AI, bringing together Europe’s leading AI platform with specialized Physics AI models for industrial simulation. If you’re working in aerospace, automotive, or semiconductors, this consolidation signals a major shift in how AI tooling will approach complex engineering challenges like real-time simulation and digital twins.

Farewell to Rails-way: When Rails-way does not work anymore — by Pawel Dabrowski

Pawel walks through the organizational and technical warning signs that your Rails app has outgrown the Rails-way conventions, from slow PR cycles and deployment friction to God objects and business logic scattered across callbacks. If you’ve felt your team’s velocity drop as complexity grew, you’ll recognize these patterns and learn why the standard Rails playbook starts to hurt rather than help at scale.

The last six months in LLMs in five minutes — by Simon Willison

Simon breaks down the major LLM developments from November 2025 through May 2026, from the November inflection point where coding agents crossed into daily-driver reliability, to the emergence of capable open-weight models running on consumer hardware. You’ll get a grounded take on what actually shifted in the space, away from the hype, using the memorable pelican-on-bicycle benchmark to track real capability improvements across models.

A New Look for Express — by Sebastian Beltran

Express got a complete redesign: new website built with Astro, overhauled documentation with AI-powered search and multi-version support, and a fresh logo reflecting the project’s evolution. If you maintain Express apps or contribute to the ecosystem, you’ll want to know about the improved docs structure and how to take advantage of better versioning and AI-assisted search.

I don’t think AI will make your processes go faster — by Frederik Van Brabant

Most organizations chasing process optimization miss the real bottlenecks by throwing bodies or AI at visible problems instead of understanding upstream constraints. Learn why detailed requirements and predictable high-quality inputs matter far more than raw speed, whether you’re optimizing for humans or AI.

Languages, Tools & Framework releases

Node.js 26.2.0 — by Antoine du Hamel

Node.js 26.2.0 brings several notable improvements including stable stream.compose, Temporal.Instant support in fs.Stats, and new HTTP 1xx status code handling. The release also hardens crypto operations, updates dependencies like undici and sqlite, and includes ongoing QUIC protocol implementation work that benefits backend developers needing modern transport protocols.

And that’s all for this week! Wish you all a great weekend and happy coding!

We want to work with you. Check out our Services page!