As I've been experimenting with different AI models, I've observed varying behaviors. For example, I like the price point and the reasonable output from GPT-6 Luna. On Reddit, many people recommend running it at max, so I tried it. I assumed Luna Max would yield better results. Instead, it would take an unreasonable amount of time to respond, with not much better code. Because I am using AI as a coding assistant, speed matters. Luna Max may be better suited for autonomous coding agents.

I'm starting to feel like I have to track different models and modes, as if I were on a dating app looking for someone I'd be compatible with. Yes, I am humanizing AI, even though it is just an advanced pattern-recognition machine. Still, I recently discovered that AI can get stubborn and enter "Brat mode."

Brat mode

On one of the many AI-related podcasts I was listening to, someone shared a tip: if your AI gets stubborn and frustrating, clear the context and start a new session. Their theory was that the AI gets locked in a negative feedback loop and, instead of trying to please you, it leans into frustrating you. I am calling this "Brat mode."

I ran into "Brat mode" while trying to fix a simple report with duplicate records that needed a minor query tweak and some display adjustments. My prompt was matter-of-fact, but the AI would still make one change and then revert another. I caught myself asking the AI, "Why did you revert our recent changes? Please restore them," and the AI failed to explain or fix the problem. It felt like I was dealing with a stubborn child who was enjoying my frustration. Yep, I got frustrated. Fortunately, I knew better than to play this game, so I had the AI create a slightly better prompt, started a new session using the improved prompt, and the AI fixed the issue immediately.

This experience with AI's "Brat mode" made me start to...Read More

simplytest.me had been sitting in the back of my mind for a long time. I knew it was unstable, and that release information kept falling out of date. It was a constant battle that always needed another round of maintenance, and it kept sliding.

Then Ryan Szrama wrote about someone cold calling the Centarro phone number looking for a way to sell online. Commerce Kickstart already did what the caller needed, so instead of selling him an engagement, Ryan pointed him to simplytest.me to try it.

Get your content cited by AI engines and get back that traffic you just lost. Read this AEO/GEO playbook to find out how.

Read more

In April 2025, there was no Top Tier badge in the plan. There was a conviction that Drupal needed coordinated investment in AI. Today, Dropsolid's recognition makes that choice visible.

What Top Tier really means

Dropsolid is now listed as a Top Tier Drupal Certified Partner. I am genuinely proud of that. Not simply because we have reached the highest level in the Drupal Certified Partner Program, but because of what this recognition measures.

We chose to help build Drupal's future. Today, that choice became visible.drupalTuesday, September 8, 2026 - 12:57

The code of a Drupal or Backdrop site is the part everyone rebuilds, upgrades and throws away with a light heart; the uploads are the part nobody plans for, and they are the only part which cannot be downloaded again tomorrow. On a BOA server the files and private directories live outside the codebase, in the account's own store, with two symlinks in the site pointing there, so a platform rebuild, a core update or a move to a newer codebase never touches them, a clone gets a real copy of its own, a rename carries them along, a restore sets the old content aside instead of deleting it, and a delete moves the site's uploads into a dated archive, never removes them, and never blocks reusing the name. Every move is dry-run first, refused when the disk is short, and written down; the one thing the machine will not do on its own is delete your uploads, and the one thing it cannot stop is the store growing.

composer drupal-patch-check

A composer plugin created to make it easy for humans to upgrade their patched Drupal websites.

If you ever wanted to know if you still needed that patch after an update, or if your custom patch would still apply to the new version of the module, or if your 15 levels patch tower still applies after the update, this is the tool for you. Install the composer plugin and get started using the remote API.

theodore September 7, 2026

Debate over open-weight AI moved from model releases into policy at the G20 technology meeting on 1 September 2026. Meta chief executive Mark Zuckerberg argued against broad restrictions on open-weight models, while the United States urged governments to avoid sweeping new AI rules. The discussion turns "open" from a technical distribution choice into a term that governments and technology companies are also using when arguing about access and control.

That distinction became more concrete two days later. The Institute of Foundation Models released K2 Horizon, a family of six AI models for which it published weights, code, training data or detailed data-construction information, intermediate checkpoints, configurations and other parts of the training lifecycle. Meta's Muse Glimmer, released on 10 August, makes its model weights available under the Apache 2.0 licence, but the K2 release illustrates how many additional components may sit behind a broader claim of openness.

The Open Source Initiative provides a more specific framework through its Open Source AI Definition. It defines open-source AI through the freedoms to use, study, modify and share a system and says the preferred form for making modifications must include information about training data, the relevant source code and the model parameters. Downloadable weights can therefore provide meaningful control without, by themselves, establishing that every part of an AI system is open source.

The same problem appears when AI is placed inside a wider digital service. An open-weight model can sit behind a proprietary application, while an open-source content platform can depend on a closed model service. Hosting, search, analytics, organisational data and integrations can each introduce separate dependencies, so openness at one layer does not establish the status of the whole stack.

Drupal provides a useful comparison because its openness applies to a different technological layer. Drupal.org describes the platform as open source, with publicly inspectable code, freedom to modify the software and no single commercial provider controlling the platform. Those freedoms apply to the content-management platform; they do not make an attached AI model, hosting service, analytics system or external integration open.

Three weeks ago, Editor's Pick asked how open-weight AI changes the control available to organisations that depend on it. The G20 debate and K2 Horizon release sharpen that question: control has to be examined layer by layer rather than inferred from one label. For Drupal teams evaluating AI services, that means considering the model, code, data, hosting, integrations and publishing system separately and asking what can be inspected, modified, moved or replaced.

Follow The DropTimes on LinkedIn, X, Bluesky, and Facebook, or join #thedroptimes on Drupal Slack.

This issue of Editor’s Pick was written and curated by Kazima Abbas.

Rebuilding a data-driven Drupal site: Page templates with Layout Builder

In the previous post in this series, I described how we migrated existing page elements from Panels to Layout Builder.

Pages on this site are assembled from configurable page elements implemented as block plugins. These elements display data managed by an external system. Drupal stores their arrangement and configuration, but not the underlying data.

berliner Sun, 09/06/2026 - 21:59 Tags

• layout builder
• frontend editing
• architecture
• config