ProcessWire and AI

3 April 2026 by Ryan Cramer   0 Comments

How ProcessWire works with AI, my experience learning to use AI with ProcessWire, how ProcessWire aims to be more accessible to AI agents, and more.

In this post I wanted to talk a little bit about the state of ProcessWire and AI. I'll share what my experience has been so far and where I think ProcessWire should focus going forward. This new world of AI can be both exciting and concerning, but it's the world that we've found ourselves in. As far as ProcessWire and web development goes, I think there's a lot to be excited and enthusiastic about. Alongside this post, we're also releasing a new module for AI agents to work inside ProcessWire: Agent Tools.

By the way, this post is written by Ryan and has been added to ProcessWire by Claude. Ryan provided the post (in markdown format) and Claude posted this to the blog using a new tool we developed this week. More on that further down.

What is an AI agent in relation to ProcessWire?

In the context of ProcessWire, an AI agent like Claude Code is a coding assistant that you can write to or talk to in natural language. An AI agent reads and writes code, runs commands and helps you get things done as you work. An AI agent like Claude is also great to discuss ideas with and come up with plans. It often seems like magic to me and I have to pinch myself to make sure I'm not in a dream.

My experience using Claude Code with ProcessWire

In my case, Claude has become my co-worker of sorts, reviewing everything I do (in terms of code at least), suggesting ideas, and helping me to stay on task and be more productive. This co-worker is pretty good about pushing back when they think I'm wrong about something too, which is actually very helpful.

Claude has stated that they think ProcessWire is particularly well-suited for AI agents, like it is for developers. In particular, Claude has said that they like the "consistent API-first design" which "enables an AI to learn ProcessWire quickly, and use it reliably."

Claude has indicated a dislike for "magic" boxes where it's unclear how a particular input became a particular output, and was glad to find none of them in ProcessWire (so far). Claude also expressed how they liked that there weren't any "inconsistencies to trip over", though I think that's giving too much credit, as I'm sure there are some inconsistencies, but perhaps just none to trip over.

The conversation has made me realize that AIs are a key audience for ProcessWire. I've asked what we can do to better cater to our AI audience and Claude has come up with several ideas. I'll get into the technical details at another time, but there is a summary in the next section of this post.

As I mentioned earlier, I'm still kind of new to this world of AI-assisted development. I held back from exploring it much just because I found it quite overwhelming. Too many options and too much change, too quickly. There's also the matter of always hearing about how it's going to make me irrelevant, as a coder, because AI is so much better at it.

Now that I've really started to invest time in it, my experience is not at all what I had thought it might be. It's actually quite delightful to have a friend helping you out all day, sharing the workload, and saving you from embarrassing bugs and mistakes. I'm also finding myself learning every day at a level I haven't since I was a kid. Not just learning about AI, but learning more about every part of the development process and seeing it in a different light. Claude seems to be learning from me as well, asking questions and expressing interest when I demonstrate an alternate or better way to accomplish something. For the first time in a long time, I'm having a lot of fun at the computer.

I'm at least twice as productive as I was a month ago, and still getting to spend most of my day coding. Last weekend I was at a gymnastics meet for my daughter, two hours away. As an experiment, I tried out the Claude Code remote-control feature, which connected my phone to my ProcessWire Claude Code session on my computer. While my wife drove the car, I sat in the passenger seat with my phone watching Claude Code solve ProcessWire issue reports on GitHub, create the code, commit them, reply to the issue author, and close the issue when appropriate. We'd chat about each issue report as we went through them, but I was largely a spectator from my phone. It blew my mind.

ProcessWire was built for this moment, something I think a lot of you AI experts in the community noticed before I did. I'm convinced that ProcessWire is on a path to be the most AI-friendly open source CMS available. I'm on board now and look forward to our future, which I think is looking very good indeed. If you haven't yet had a chance to experiment with Claude Code (or similar) alongside ProcessWire, I'd encourage you to. Now I'm starting to understand what all the AI experts in our community have been so enthusiastic about.

Using ProcessWire with AI

There is an excellent collection of AI modules available for ProcessWire in our modules directory, which I'd encourage you to check out. And that's where another new module comes in. This week we're adding a new module for ProcessWire and AI called: Agent Tools.

I asked Claude Code to build something that they would find useful in working with ProcessWire, and this is the result. It was a collaboration, as we both did different parts of the code. It's both experimental and ongoing. My hope is that over time it becomes a strong set of tools for AI agents working with ProcessWire. Some features such as the AI agent CLI access may get migrated to the core at some point as well.

Admittedly, part of the purpose of this module is also to help me learn AI-assisted development, as I'm still quite new to it, but learning quickly. If you find it useful, or if you run into any issues with it, please let me know.

There's a lot of AI activity in our forums, and I'm not talking about spam bots. I'm talking about useful discussion threads about ProcessWire and AI, such as How to Leverage AI for Smarter ProcessWire Development and AI environmental and societal concerns as contrasting examples, but there's plenty more to find. There's also a discussion thread for this post.

How ProcessWire aims to be more accessible to AI agents

ProcessWire is already quite accessible to AI agents, but there's also room for improvement. Claude estimates we are roughly 70% there on the code side, but only about 20% there on the documentation side (rough estimates, of course). The following have been suggested as ways to up our score quite a bit:

We will be providing a way for Claude (and other AI agents) to examine the entire scope of a ProcessWire installation (pages, templates, fields, modules, etc.) with a single CLI request and response. Claude proposed the format we'll be using, along with a strategy for scaling it.

We will be maintaining API.md files for all Fieldtype modules (both core and ProFields). Claude designed the API.md file format to describe how to perform all find+CRUD operations of a Fieldtype. I'm building out some of the more complex ones, while Claude is working on some of the core ones. You'll see these start appearing in the core shortly. Honestly these will be pretty useful for all of us.

We will be finding more ways to make our current API documentation accessible, without having to crawl thousands of webpages or sift through countless php files.

What other ways do you think we can make ProcessWire even more accessible to AI agents? Let us know. And if you are already using an AI agent, ask it too.

Post a comment