Welcome back. Last week, OpenAI signed a deal with the Pentagon, and many users aren’t happy about it. A massive wave of ChatGPT users switched to Claude (led by public figures like singer Katy Perry). The wave of defections was so big that Anthropic's servers couldn't keep up. Claude ended up going down for hours, hit by one outage after another.

Also: How to fix AI coding sessions that spiral, why the fastest devs in 2026 moved away from MCP, and a local model that runs entirely on your phone.

Anthropic's Claude hit by back-to-back outages: The AI lab is dealing with rolling outages after its Claude chatbot went down, taking down services across the platform. The surge was reportedly driven by ChatGPT users switching over after OpenAI's controversial Pentagon deal, and the rollout of a free Memory feature. Anthropic also announced that voice mode is rolling out for Claude Code.

Alibaba launches small AI models that rival ones 13x their size: The Chinese tech giant Alibaba just dropped four open-source Qwen 3.5 models with impressive efficiency gains. The 9B model matches OpenAI’s GPT OSS 120B in key benchmarks while being 13x smaller, and the 4B version variant rivals the previous 80B version. The smallest variants (0.8B and 2B) are small enough to fit on your phone, which means they work offline with no cloud dependency. You can run the model locally with just ~7GB of space.

Apple unveils two $599 devices built for on-device AI: The tech giant just launched the iPhone 17e and a refreshed iPad Air, both designed to run Apple Intelligence locally. The iPhone 17e ships with the A19 chip and doubles base storage to 256GB, while adding AI tools like Call Screening and Live Translation. Meanwhile, the iPad Air gets the M4 chip with 50% more unified memory, giving AI workloads significantly more room to breathe. Pre-orders open March 4.

Move from idea to production faster with a platform designed to scale with you. Get access to dedicated engineering support and technical guidance to make sure your architecture is solid from day one.

Build on a unified data platform that brings together document storage, Vector Search, and built-in scalability. Simplify your stack. Ship faster with fewer tools.

Jumpstart development with MongoDB Atlas credits. 

Join thousands of startups representing $200B+ in combined valuation.

Anthropic drew two red lines. CEO Dario Amodei refused to give the Pentagon unrestricted access to Claude, holding firm on two points: no mass surveillance of Americans and no fully autonomous weapons. The White House struck back by labeling Anthropic a supply chain risk to national security — a designation usually reserved for foreign adversaries. Anthropic has announced plans to fight the designation in court.

OpenAI jumped right in. Hours after the government told agencies to ditch Anthropic, OpenAI locked in its own classified deal with the Pentagon, promising similar safety guardrails. Sam Altman called the Anthropic ban a "very bad decision," while admitting his own deal didn't exactly look great given the timing.

Meanwhile, the research paints a worrying picture. A King's College London study put GPT-5.2, Claude Sonnet 4, and Gemini 3 Flash through 21 simulated war games — 329 turns, nearly 800K words of reasoning. In 95% of scenarios, at least one model deployed a tactical nuclear weapon. Not once did any choose to surrender.

The backlash. The Claude app climbed to the top of the App Store as a "Cancel ChatGPT" movement gained traction across X and Reddit. Over 90 OpenAI employees and 600 Google employees signed an open letter backing Anthropic's position. It's the first time a policy stance has triggered a measurable shift in developer loyalty — and likely not the last.

No clear answers. While Anthropic’s objections seem reasonable, a bigger concern looms: in a democracy, who gets the final say on such matters — elected government officials or CEOs of private companies?

Claude Code performance tends to drop the longer you use it. The context gets messy, responses feel sluggish, and it eventually loses the thread of what you were discussing just minutes ago. Ex-Google software developer YK Sugi pointed out a solid workaround using Claude Code's built-in plan mode.

Just hit “Shift+Tab” to jump into plan mode and have Claude pull together everything needed for your next session. Since the next agent starts with a completely blank slate, you'll need to be thorough:

Claude will then scan the codebase, collect relevant files, and build a plan. Once finished, you'll see this output:

Go with option 1. This clears out the old conversation and kicks off a fresh session with only the plan loaded.

How to set up a local AI coding workflow: You’ll learn how to build a powerful local AI coding workflow using LM Studio and Claude Code. The tutorial walks you through linking local models across devices, optimizing context windows, and deploying AI sub-agents to build a full-stack Next.js app while keeping data private.

Antigravity Awesome skills: A curated library of 1000+ agentic skills built to work with every major AI coding assistant, from Claude Code to GitHub Copilot. It's a solid starting point for devs who want to turn their AI tools into something closer to a full-stack digital agency.

Everything is Context: Current AI systems struggle to organize and track the external knowledge and memory they need for reasoning. To solve this, researchers developed a file system approach that turns scattered AI context into a permanent, traceable, and easily managed pipeline.

Grow customers & revenue: Join companies like Google, IBM, and Datadog. Showcase your product to our 200K+ engineers and 100K+ 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