Cloudflare vows to stop breaking the internet

The Payload team released version 3.69. The crowning feature of this version is the ability to use modular dashboards and widgets. I think I'll take a look at this feature in more detail in a live stream next week. Be sure to check that out for more details on how this beta feature works.

The team added some new AI development resources to their templates. All templates now have `AGENTS.md` and `.cursor/rules/` by default. This should improve how your favorite AI tools interact with fresh Payload projects.

Payload released a couple of key bug fixes as well. First, the MCP plugin now adds the `collectionSlug` and `_strategy` to the `user` when using MCP operations. This fixes an issue where the `user` came through as undefined in MCP operations.

The second key bug fix was in the S3 storage adapter. This is a simple fix that uses `encondeURIComponent` to ensure the plugin generates properly formed URLs.

There's been some discussion in Discord and on GitHub about using TanStack for Payload. It's early on in the discussion, but a few members of the community are working on creating a TanStack Start core for Payload CMS. I'm excited for the doors this opens for Payload as they become more and more flexible.

If you're having issues with keeping your data up to date in Payload CMS and your frontend, this video walks you through how to keep everything in sync using revalidation and Payload hooks.

Since the release of version 16.1, the Next.js team has focused on making their tools more reliable by releasing version 16.1.1 and several canary updates. They fixed specific issues with how Turbopack handles files on Windows and updated the code to stay safe from newly discovered security risks. They also polished the new "Bundle Analyzer" tool to help developers see exactly what makes their websites slow or heavy.

Cloudflare has declared "Code Orange." Basically, they want to stop breaking the internet. To accomplish this, Cloudflare will start testing setting changes slowly instead of sending them out to everyone all at once. This is also part of a modernization push to get their emergency tools fixed and in better working order.

This effort should keep most apps and websites working, even if small parts of Cloudflare's systems break. You'd think that this would've been their plan from the start. A soft and slow rollout of new features just seems like a good idea when you're a large corporation.

The internet experienced another major outage on December 24th. No one is taking responsibility for this one, but signs point to AWS. They've since denied it, but it seems that users in the US and India reported issues accessing Fortnite, Arc Raiders, and other games.

As of the time of this recording, the status page for AWS appears to be all clear, so I don't think we'll know what happened with this one for a while.

An investigative reporter is leading a group of authors to sue essentially every AI company for using copyrighted books without permission to train their models. Perplexity claims they did not and have not used books to train their AI, and this follows an already settled claim against Anthropic. This is the first case, though, that names xAI as a defendant.

The suing group is not organizing in a class action lawsuit, similar to what we've seen in previous lawsuits. The key difference here is that if AI companies want to settle, they'll need to settle with individual authors, not a class of authors. That means settlements will most likely be larger and spread across fewer people.

I'm interested to see where Anthropic ends up in this, as they've already settled a large class action lawsuit. Saying a settlement didn't go far enough to protect authors is not really a defense, but it is within the rights of authors who did not join the class action lawsuit to sue on their own.

This will likely drag on in court for years, but I'll do my best to keep up with the news.

OpenAI plans to put ads inside ChatGPT to earn more money. I mean, they have to, right? How else will they turn a profit this century? I just hope they label these clearly, otherwise it's going to be impossible to trust any answers that ChatGPT gives us. Trust will erode fairly quickly for the more technical users, while others may take a while to recognize what's happening.

On the other hand, developers and marketers will have another paid channel to target. They'll be able to show products that—ideally—answer questions to people using ChatGPT. Ads always find a way into platforms. Let's hope ads stay in only the free plan.

Attackers continue to hit OpenAI's Atlas browser with prompt injection attacks. So, their team is using AI to attack its own system. The goal of this exercise is to find and patch vulnerabilities before a bad actor gets to it.

Hidden text on websites can trick AI agents, and this isn't an easy problem to fix. You'd think that companies would attempt to fix this issue before launching a product. It just goes to show that these companies would rather be first to market than safely in the market.

Cursor released its last update of the year. First, it's now easier to customize your layout across workspaces. I primarily use JetBrains products, so it's been difficult for me to understand how to use Cursor's interface. Maybe this will help me understand it better.

Other than that, they released some stability improvements and fixed some bugs. Not a very exciting update, but it was a very necessary update.

Google released a few new Gemini features this month. First, you can draw or type onto an image to prompt Nano Banana to make precise changes to an image.

Second, Gemini now allows you to bring NotebookLM notebooks into Gemini. This new feature means you can have better context and fewer hallucinated responses from your source materials.

The last new feature I'll cover is a new way to use Gemini's Gems. Gems are personalized, custom AI assistants in Gemini. Google is testing a beta feature that allows you to use Gems to create interactive apps for your tasks.

Railway unveiled a new home page. According to the team, this change attempts to explain what Railway does for you instead of you having to fumble around for the words. I have to say, it does look nice

After releasing railway dev last week, the Railway team released a new terminal user interface, or TUI. This interface provides tabbed service logs, service information shown at a glance, and keyboard-driven navigation. Developers rejoice.

The service logs provide all relevant information for all your services in one place. They're organized neatly into tabs for easy navigation. The service information shows status information, such as the local URL and number of environment variables loaded. To navigate through all this new information, the Railway team implemented keyboard navigation, so you don't need to scroll through anything with your mouse. It's meant to be as convenient as possible.

The team also released a high-availability Postgres template and template metrics. The Postgres template is available now and uses a Patroni-based Postgres cluster. This provides automatic failover and load balancing out of the box. It's simple to set up, so you now have this option if the default Postgres options don't cut it for you.

Template creators can now see how their templates are doing. You can see how many deployments there are, what usage looks like, and—most importantly—how much you're earning from Railway's template kickback program.

Railway had a busy year this year, and I'm looking forward to what they implement next year.

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

You're now able to bring your own keys to JetBrains IDEs. This allows you to use your preferred models, locally hosted models, and more. Even better, you don't need a JetBrains AI subscription. Your machine stores and manages your keys and is never shared with JetBrains.

JetBrains will continue to add more providers, such as Gemini, Azure, and Bedrock. Currently, JetBrains only supports Anthropic and OpenAI API-compatible providers.