Welcome back. Should you hustle 24/7 as a dev, or should you protect your work-life balance? After Corgi's founder (somewhat controversially) claimed that you're bound to lose if you aren't working seven days a week, Linear’s CEO broke down why that’s not actually the secret to winning — read his reasoning. DHH, creator of Ruby on Rails, had similar thoughts.

Also: How to set up your codebase for self-improving agents, track everything your AI builds with one prompt, and see what an ex-Google staff engineer does first at a new job.

Nvidia unveils America's most capable open model: The chip giant just announced Nemotron 3 Ultra, the most powerful open-weight model made in the U.S. Its massive 1-million-token context window means agents can process an entire codebase in one go. Nvidia claims it’s up to 5x faster and 30% more affordable than competing open models, and they’re releasing the weights and training recipes publicly. One caveat: China's Kimi K2.6 still holds a solid lead in raw intelligence. It officially launches on June 4th.

Alibaba ships a new model that writes and tests its own code: The Chinese tech giant just dropped Qwen3.7-Plus, a multimodal model designed to take action instead of just providing answers. It can process images and video, then handle the entire coding loop from start to finish on its own. The model writes its own code, calls external tools, and keeps testing until the job is complete. It runs on the Bailian cloud platform, which includes guardrails to keep autonomous agents within their limits as they execute commands and edit files.

xAI brings Cursor's Composer 2.5 to its coding agent: The Grok team just upgraded their coding agent, Grok Build, with Composer 2.5. It’s a fast, state-of-the-art model that excels on long-running tasks and following complex instructions. The model is available now for SuperGrok and X Premium+ subscribers. Try it by typing /model and then selecting Composer 2.5 in the menu.

Your dev stack got an AI upgrade everywhere except the input layer. You're still typing every prompt, every ticket, every review comment by hand.

Wispr Flow closes that gap. Dictate into Cursor, VS Code, Slack, Linear, or anywhere else you work. It's syntax-aware: camelCase, snake_case, acronyms, and file names all come through clean. Mention a file in Cursor or Windsurf, and it auto-tags.

It's the voice layer for an AI-native workflow. Speak your intent. Your tools do the rest.

Available on Mac, Windows, iPhone, and Android. Used by millions of developers, including teams at OpenAI and Mercury.

Try free

The enterprise subsidy quietly ran out. This spring, Anthropic and OpenAI finally pulled the plug and moved their biggest customers off deep discounts onto list pricing. They now have a product worth paying full price for. Django co-creator Simon Willison calls it product-market fit. And the fit is your engineering team.

Chatbots got famous, but they never paid the bills. Hundreds of millions used ChatGPT for free without ever upgrading. Then coding agents arrived. They put a token-hungry tool in the hands of the most expensive employees in the building, and engineers started running them all day long. When demand is that high, a price increase doesn't slow anyone down, so the labs stopped offering discounts.

The payoff showed up fast. Anthropic just turned its first profit, doubling revenue in a single quarter. But that profit is a bill on someone's desk. Microsoft rolled out Claude Code to thousands of engineers in December, but token costs torched its annual AI budget in just months. So the company is pulling the plug on the tool by June 30.

The ball is in your court. AI labs won't subsidize you forever — it’s up to you to do the math on what it’s worth to your team. With power users spending up to $2,000 a month (and a lot more in some cases), forward-thinking teams are tracking ROI by tying costs to PR throughput rather than vanity metrics like AI-generated code percentage or tokens consumed. This guide offers a solid baseline to turn these costs into a manageable line item.

Connect AI agents and coding assistants to your email stack without giving up control.
Mailtrap lets teams:

Start sending safely with Mailtrap

Cursor often confidently imports packages that aren't even installed and sometimes ones that don't exist at all. A study of 576,000 AI-generated samples found that nearly a fifth of the recommended packages were fake, and attackers are already registering those made-up names with malware.

A simple Cursor rule prevents the agent from trusting its own memory regarding your dependencies. Create a rule file at “.cursor/rules/verify-deps.mdc” and set it to always apply.

Cursor now verifies that a package actually exists before writing the import. This way, any hallucinated dependencies are caught immediately rather than breaking your next build.

P.S. Get 50+ AI coding hacks for Claude Code, Cursor, and Codex here.

How to save 90% of Claude Code token usage (by Meta Staff engineer): This tutorial teaches developers four strategies to reduce Claude Code token usage by up to 90%. You will learn how to pre-index codebases, compress noisy logs, shorten AI responses using Caveman, and optimize session management to balance cost and quality.

OpenStatus: An open-source platform that handles your uptime and synthetic monitoring, tracking APIs and websites from all over the world. You can easily spin up a public status page to show off your reliability.

Hermes Web UI (11.8K ⭐): The easiest way to access Hermes Agent via web or mobile. Built entirely with Python and vanilla JS, it requires no complex frameworks, bundlers, or build steps.

Rethinking search as code generation (by Perplexity): Traditional search is a bottleneck for AI agents. By treating search as code generation, we can create programmable building blocks that allow agents to build custom retrieval pipelines, seriously boosting efficiency and performance.

This tutorial covers the 6 best skills that will level up your dev workflow.

Grow customers & revenue: Join companies like Google, IBM, and Datadog. Showcase your product to our 300K+ engineers and 150K+ followers on socials. Get in touch.

You can also reply directly to this email if you have suggestions, feedback, or questions.

Until next time — The Code team