Apple gives devs a folding canvas to build for

Welcome back. Remember the SpaceXAI engineer casually shipping 2,000 PRs a month? She just dropped Part 2 of her playbook, and it just might be even better than the first. Meanwhile, Apple just launched their first foldable phone, and iOS devs are already mentally redesigning every app they’ve ever built.

Also: The ultimate GrokBot guide, more Anthropic employees are weighing in on the 'AI ending humanity' debate, and scrolling your MacBook by hand gestures.

Apple gives developers a folding canvas to build apps for: The tech giant just unveiled iPhone Duo, which gives iOS developers a new hinged form factor to design around. Apps will need to adapt across fold states, split view, and shifting orientations. The iPhone 18 Pro also adds a new feature called Reference Image, letting you check for AI edits by comparing photos to their original digital negatives. Plus, iOS 27 drops next week with a new Siri powered by Google’s Gemini models.

SpaceXAI engineer drops her secret to shipping 2,000 PRs a month: Lauren Tan, whose guide we shared last week, is back with Part 2 of pstack, her skill set for doing rigorous work with AI agents. While part 1 was all about verification, this time, she shows how she plans and prototypes while keeping quality high at scale. Her advice: skip bloated plan docs that only look like progress. Instead, work in small, verified changes you're happy to throw away when they fail. Try pstack here.

HuggingFace introduces an autonomous ML engineer: The model hosting platform just unveiled an ML intern for you inside their chat window. You can start with a plain-English chat, and it handles the research, dataset building, and model training. It even uploads ready-to-use demos to the Hub. Every run gets its own dashboard so you can track progress, and you can also set a compute budget that it strictly follows. Give it a try. 

AI is moving beyond copilots. At {unscripted} Virtual, leaders from Anthropic, SpaceXAI, Workday, Riot Games, Google, and more will explore what happens when agents plan, build, test, secure, and ship software.

Hear how teams are designing the SDLC for more autonomy without giving up trust, governance, or developer velocity.

September 30 | Virtual | 10:00 AM PT / CEST

Register Now

Software development just forked. Nearly every engineering team has coding agents now. They write code, run tests, and fix failures for hours without a human. Yet, most teams aren't shipping any faster. The real split is between developers who just changed their coding tools and those who changed how they work. Only the second group is seeing massive productivity gains.

This insight comes from Clare Liguori, a senior principal engineer at AWS. She argues that top developers have stopped building software directly. Instead, they build the agentic systems that build the software. Once that setup is ready, they manually write less than 1% of the code they ship. Liguori calls this frontier engineering. She just dropped a guide to help other developers make the switch.

Your new job description. Her guide covers 10 core principles. Read together, they come down to three jobs, and those jobs are how you spend your day once you move beyond manual coding:

The payoff compounds. Those first few weeks feel slower. But once you get through them, the re-architecture you kept putting off and the tests you never had time for finally click into place. You can find the full guide here.

How a senior dev ships 50+ PRs a week while on vacation: You’ll learn how to run multiple coding agents in parallel, automate code reviews, and use dedicated dev environments to keep work moving without staying glued to your machine. The tutorial also covers how to build better safety nets around merges and turn recurring workflow friction into tooling your agents can handle.

Traycer: A shared workspace for your coding agents. You can run Claude Code, Codex, OpenCode, and Cursor side by side, let agents hand off work across chats, and keep context, artifacts, and history in one place.

How to cut Gemini token use by 76% (by Google): The search giant shows how to put a context-compression layer in front of your agents to strip junk tokens from tool outputs before they hit Gemini. In its tests, Headroom cut prompt tokens by 76% with no loss in task accuracy. This guide shows how to wire it into Gemini 3.8 Flash, Google ADK, and OpenCode, plus where compression can break down.

You pay for Claude Code, Codex, and Cursor, but you can only drive one at a time, the rest of your quota sits idle. An AI builder shared a fix: install them all on Grok Bot's cloud machine and let it delegate.

Diffs and test results come back to your chat. Repeat the same setup for Claude Code or Cursor.

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

Senior dev Matt Pocock just shared the 5 skills that matter most in the AI era, including whether beginners should still learn to code and what senior devs need to get better at now.

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.

We put together a few guides on coding agents, agentic engineering, and leadership frameworks to help you level up in your career. Browse all our guides.

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

Until next time — The Code team