In partnership with
Scale AI support on AWS, see how July 9
Customer expectations keep rising. Support budgets don't. On July 9, Fin and AWS are hosting a live executive session on how leading enterprises close that gap: scaling AI-powered support while simplifying how they buy it.
You'll see how to resolve an average 76% of conversations with Fin on AWS enterprise-grade infrastructure, procure through AWS Marketplace to put committed cloud spend to work, and turn the Fin and AWS collaboration into lower support costs. Register for the live session to see how.
What's up, everyone? This week the government became AI's most powerful player: Anthropic got its models back while OpenAI volunteered to hold its own back. Plus Sonnet 5 arrives, and Cloudflare wants your agents to pay per request. Let's dive in.
You currently have {{rp_num_referrals}} referrals.
Anthropic got its models back, but the government kept the keys. On June 30, Commerce lifted the export controls it placed on Fable 5 and Mythos 5 two weeks earlier, and both are rolling out again. Anthropic is positioning it as a partnership: a new industry framework for scoring jailbreak severity, plus expanded pre-release access for government evaluators. Claude being back counts for less than the new normal, where Washington expects a look at frontier models first and Anthropic is helping build the process that makes it routine.
OpenAI put itself on the same leash, on purpose. The same week, OpenAI previewed GPT-5.6 (Sol, Terra, and Luna), then limited the release to vetted partners at the government's request, with no binding order forcing it. The models are a step forward, with a new max reasoning effort and an ultra mode that spawns subagents to accelerate complex work. OpenAI reportedly went further, offering the U.S. government a 5% stake in the company, worth around $42.6 billion, to ease the pressure. When an AI company complies before it's told to and offers Washington equity on top, the "voluntary" review the AI executive order wanted just becomes the default.
Sonnet 5 gets most of the way to Opus for less. Anthropic released Claude Sonnet 5 as the new default for Free and Pro, and it plans, uses tools, and checks its own work unprompted. Its performance lands close to Opus 4.8 at a lower cost. If you lean on Opus in Claude Code out of habit, you have until August 31, while the introductory pricing lasts, to test whether you still need to.
Cloudflare wants your agents to pay per request. The new Monetization Gateway lets you charge for any page, API, dataset, or MCP tool behind Cloudflare, settling in stablecoins over the x402 protocol. Agents are becoming the web's main visitors, and they won't subscribe, so you bill them per call instead. It's a clever answer to a problem, and also a bet that every paid request should move money through crypto behind the scenes, which is a big dependency to build your whole business on.
shadcn finally does chat. The latest release adds a set of chat-interface primitives that handle the streaming and scrolling behavior nobody enjoys building. A new @shadcn/react package offers them headless, so the interaction logic stays tested in one place and you bring the styling. It doesn't replace AI Elements, but if you're building a chat UI, it's the fastest place to start.
Railway now meets you at the SSH prompt. Railway's newest changelog introduces Railway over SSH: run ssh railway.new and it signs you up, creates a sandbox, and connects you to the Railway Agent with nothing installed. Every Sandbox now comes with six coding agents preinstalled, and private networking is scriptable from the CLI. This one is really about agents. Yours already has SSH, so it can provision real infrastructure without an MCP server or a browser login standing in the way.
Payload patched a CVE. Payload 3.85.2 is a maintenance release, but it resolves a high-severity vulnerability in the js-cookie dependency, along with fixes for reordering that could unpublish newer drafts and duplicate IDs in nested Mongo relationship queries. No new features this time, but the security patch alone makes it a same-day upgrade if you run Payload in production.
Want to learn more about Payload and how to apply it to your project? Check out my course, Payload Essentials. The course walks you step-by-step through a full Payload and Next.js project so you can go from zero lines of code to a production website ready to deploy.
I believe high-quality education should be as accessible as possible to everyone, no matter where in the world they live. So, purchasing power parity is built in to the price. If you don't see a banner with a discount and believe you should, please let me know!
Also this week:
Google capped Meta's Gemini access. Per an FT report, Google couldn't supply all the Gemini capacity Meta wanted and has told staff to spend tokens more carefully. Even the giants can't get enough compute.
Tailwind CSS 4.3.2 is out. A maintenance release: bare spacing values for
auto-rows-*andauto-cols-*, several watch-mode and HMR crash fixes, and better class detection. All quality-of-life.
RECENT VIDEO
AI writes more of our code every week, so it's fair to ask whether learning to code still pays off. Here's my take.
What did I miss? There's so much happening in modern web dev that I know I missed something. Please share your thoughts in the comments or reply to this email. I want to address your suggestions and may include them in future newsletters.
Thanks for reading. See you next time.