GPT4 is retiring

The Payload team released version 3.73 in the past week. This version includes the long-awaited Next.js 16 support. I still haven't moved over to Next.js 16 yet, but I haven't seen much need for my projects. I'll be diving into what's new in Next.js 16 in the coming weeks, though. The team won't update templates until the Next.js team releases version 16.2. Remember, there is still no support for cache components.

If you're using SQLite, you can now enable Write-Ahead Logging (or WAL) for improved concurrent read and write performance. You opt into this by adding it to your sqliteAdapter. Also in SQLite, you can now set the max wait time when the database is locked. This prevents errors that are common in high-concurrency situations.

For databases managed by Drizzle, the Payload team released a command to migrate your existing project to use blocksAsJSON in your database adapter. This command updates your config and generates the migration you need to pull this off. blocksAsJSON can help your database perform better when you have many blocks to choose from.

You can now use a requestHandler argument in your useLivePreview hook in React and Vue. This allows for the customization of how you fetch your data. If you use frontend proxies or external middleware, this should help you out.

The MCP plugin now supports the select API. This allows you to reduce your token usage by selecting only what you need from a Collection or Global.

The last new feature is that the Payload team implemented React's useEffectEvent whenever it's available. If you're on an older version of React for any reason, the legacy versions will fallback to the existing method.

For bug fixes, you can now use duplicate values in select fields with hasMany set to true. I didn't realize this was a bug. I actually wanted to use this for one of my live-streamed projects. I might have to revisit it now.

Outside of the release, the Payload team released a new skill for Payload. I've tried it out with some good results in Claude Code. I'm interested to hear your perspective, though. It's at about 80% of where it can be, but I expect it to get much better in the near term.

I took some time to explore how to change up the design of your Payload emails using React Email. Let me know what you think!

Figma had a busy week. First, you can now use FigJam diagramming in Claude. This is great for quickly mocking up flow charts and other diagrams.

They've brought their "glass effect" tools out of beta and into general availability. Now you can add a glass effect to anything and pretend you're an Apple designer. But seriously, this is great for those who are designing apps for the Apple ecosystem to see how their product fits into the whole picture.

Lastly, Figma is available in Google Chat as a Google Workspace add-on. I have Google Workspace. Where is Google Chat? Does anyone use it?

Anyways. Shadcn updated Base UI's components to support inline-start and inline-end side values. 8 components now support the values, so feel free to update the tooltip, popover, combobox, context menu, dropdown menu, hover card, menubar, or select Base UI components. Or ask your favorite LLM to do it for you.

Someone discovered yet another security vulnerability in Next.js. Attackers could deny service to your site through a vulnerability in React Server Components. I'm getting tired, friends. This one impacts versions 13 and up, so be sure to update to the most recent stable patch of the version you're on to stay safe. I'll be going up to 15.4.11.

Other than that patch, the team released multiple patches backporting bug fixes to older versions of Next.js. They updated the 16.2 canary branch a few times, which addresses memory leaks, cache lifecycles, and more.

OpenAI is retiring the last model I enjoyed using of ChatGPT. They're retiring GPT-4 and older versions with claims that GPT-5.2 does the same stuff but better. I have yet to see that. These models are still available via API, but I expect that should change soon. Now if only they could fix the word vomit that ChatGPT churns out. That dude is verbose.

OpenAI also launched Prism, which is a free workspace for scientists to write research papers with GPT-5.2 as their research assistant. It's meant to be a one-stop shop for all the tools scientists and their teams need to do their research.

Cursor isn't neglecting their IDE quite yet. They released several quality of life updates to both the editor and the CLI.

First, the editor now supports subagents, which handle smaller parts of a parent agent's task. They can run together within their own context. You can also configure them with custom prompts, tools access, and models. This is meant to speed up the overall execution of your agents.

Cursor now supports Skills in both the editor and the CLI. I haven't gotten Payload's new skill in my Cursor editor yet, but I also haven't tried very hard. Either way, your Agents can discover and use skills when specific knowledge is needed. Skills are used via slash-command.

You can now generate images from Cursor's agent using Google Nano Banana Pro. You'll see these images as an inline preview, and they'll be saved to an assets folder by default. This makes it easier to create UI mockups without sourcing placeholder images.

Similar to Claude Code, Cursor's agents now ask clarifying questions in the form of a survey. I love this feature, as it helps me understand what my agent needs to know before starting to code.

Anthropic signed two major enterprise deals this week. They will work with ServiceNow's to integrate Claude into its workflow platform. The other deal is with the government of the United Kingdom. The UK will use Claude as its AI assistant for its government services. This brings an enterprise-caliber agentic system to the people of the UK to use for government assistance.

The Anthropic team builds fast when it comes to Claude Code. They released a slew of updates from version 2.1.16 to 2.1.25. These updates are primarily minor bug fixes that support Claude Code as a high-performing CLI tool.

Notable highlights from the update include native plugin management support and OAuth for VS Code, customizable keyboard shortcuts by running /keybindings, and fixing parsing errors in JSON.

Google is integrating Gemini into Chrome to power its new tool, "auto browse." This is essentially OpenAI's Atlas, which is a Chromium browser, but with Google and in Chrome. This is much more likely to reach the masses, so it might be time to think about how AI assistants see your site if you haven't already. This is especially true if you're in e-commerce.

Gemini will use the new "Universal Commerce Protocol" to let its AI agents perform transactions. This protocol is a set of structured data that you will need to provide if you want AI agents to interact with your site in a seamless way.

Railway released a new view to your projects screen. Instead of only having a list view of your projects, you can now see what architecture is contained in those projects at a glance.

They also released focused environments for monorepos and multi-service projects, but there's not much to write about this as it's still in beta. This simply allows you to focus your deployments on only the services you need to redeploy. If any service is dependent on another, that service will also be redeployed leaving the rest untouched.

Use my affiliate code to sign up for Railway if you want.

The Astro team released version 5.17. This version allows you to configure the dev toolbar placement and adds support for partitioned cookies. These changes aren't very exciting, but they do give you more control over the behind-the-scenes stuff that causes headaches for your Astro site.

You can now choose what background color you want your image to have when you transform images that don't support transparency. This used to default to black, but you can now choose any color you'd like using Astro's image tools.