FileMaker AI development is changing what’s possible inside a point-and-click environment. FileMaker is not a folder of text files, so it can feel as though there’s no obvious way for AI to reach in and create real FileMaker schema — no door in the wall.
But there are doors. You just have to know where they are.
The Door in the Wall, with an introduction by Ronnie Rios. Recorded at the Full Access Unconference, Los Altos, August 7th, 2026.
For more context on the platform and tooling mentioned in the session, see Claris FileMaker, the FileMaker OData API Guide, and the FileMaker Upgrade Tool Guide.
Highlights and Demo Takeaways
Ronnie Rios opens with an excellent short introduction on how our work as developers is evolving as AI takes on more of the building — and on the questions and the anxieties that come with that.
The three doorways. Before the demo, Ivan lays out three practical routes for FileMaker AI development and how AI can reach into FileMaker schema.
- OData is the narrowest — tables and fields, handy for specific jobs.
- Paste, the clipboard method, is much more powerful: FileMaker already turns copied schema into XML behind the scenes, so if AI generates that XML directly you can paste it in as though you’d copied it from a real file. It works with almost everything you can normally copy and paste, but that means it will not work for relationships, because you can’t copy and paste on the relationship graph.
- And Patch, built on the FM Upgrade Tool, which handles all of FileMaker’s schema, relationships included, and can run in an agentic loop. With this method, you can construct the XML file that defines FileMaker schema, apply it to a file, read the file back to confirm it landed, then build the next layer on top.
These techniques aren’t in competition. I think of them as tools in a tool belt, and more than one may be useful on the same project.
The live demo. Ivan uses AI to take an empty FileMaker file and turn it into a working invoicing solution in about six and a half minutes in front of everyone. It is a practical example of FileMaker AI development in action: tables, fields, relationships, calculations, portals, a custom theme with layout object styling, and a button that emails a properly formatted invoice to a customer as a PDF. He gives an initial prompt, then stands back while the AI does the work.
(It’s worth acknowledging that Ivan is not the only one working in this territory. Wim Decorte has spoken publicly about his excellent work along the same lines, and what you see here is Ivan’s own flavor of the Patch technique.)
The session also explores why themes are so difficult for AI to generate, and explains the two tricks that make it possible with this technique. This was a fairly recent breakthrough for Ivan at the time of the talk, so he spends a few minutes nerding out about the details.
The case for native FileMaker layouts. AI-generated web viewer UIs are stunning and we’re all going to be using them more. But transaction control, Find mode, sorting, print integration and layout-driven security all come free in native layouts and have to be rebuilt if the UI moves into a web viewer. I argue that both UI approaches are important and should exist side by side.
Where this leaves us. Wrapping up, Ivan offers some of his thoughts on the larger questions raised by Ronnie in his intro, like how our roles are shifting as these tools become more capable, and why the changes play to strengths FileMaker developers already have. The real advantages are human relationships and architectural knowledge, not simply the ability to write code.
Q&A
Some great questions from the audience, including:
- Do you still have to QA what AI builds?
- What happens when something goes wrong? How does the AI handle it?
- Does the Patch technique work on hosted files?
- How much upkeep do the skills files really need?
- Could any of this rescue an old classic layout?
Stay curious!
Chapters
- 0:28 — Ronnie Rios: great development was never about writing code
- 5:26 — the door in the wall: why FileMaker feels closed to AI
- 6:40 — Doorway 1: OData
- 7:05 — Doorway 2: the Paste / clipboard method
- 9:50 — Doorway 3: Patch, built on the FM Upgrade Tool
- 11:14 — closing the loop: Save as XML, code in / code out
- 13:16 — the harness: Claude Code, AI skills, and my own flavor of this
- 15:26 — the design brief: what I decide vs. what AI decides
- 17:47 — the file, genuinely empty
- 18:13 — the initiating prompt, spoken
- 21:17 — Cycle 1 · tables and table occurrences
- 22:30 — Cycle 2 · every field
- 23:32 — Cycle 3 · relationships, which Paste cannot do
- 24:54 — why JSON first: decisions to AI, exact code to a builder
- 25:10 — Cycle 5 · the Email Invoice script
- 26:19 — Cycle 6 · layouts and a custom theme
- 27:57 — why themes are genuinely hard
- 29:27 — capturing a known-good theme as a foundation
- 31:19 — the trick: a new theme needs a layout attached, or it vanishes
- 33:01 — the relationship graph it drew
- 33:54 — the custom theme, and editing a named style live
- 35:43 — the agentic loop, in depth
- 37:46 — the AI skills behind it, and putting your own conventions in them
- 41:11 — the FM Upgrade Tool’s verification blind spot
- 44:10 — you always QA. Always.
- 45:53 — editing existing schema, and the reuse-the-original-ID workaround
- 47:58 — tools in a tool belt
- 50:27 — the case for native layouts
- 52:30 — inheriting AI-written web viewer code as an archaeology project
- 55:15 — where this leaves us as FileMaker developers
- 56:44 — the role shift: thinking like a tech lead
- 59:05 — our real advantage, and why clients still call us
- 1:00:22 — Q&A begins
- 1:02:50 — hosted files: not yet
- 1:03:30 — how much upkeep do the skills files actually need?
- 1:06:05 — capturing every XML variation
- 1:08:17 — what’s an acceptable speed?
- 1:11:02 — could Patch re-theme an old classic layout?
More FileMaker AI Development Resources
This session draws on the FileMaker AI Foundations series Ivan has been writing on the blog. If you’d prefer the written versions, in reading order:
The two most directly connected to this session are The Paste Method and The Patch Method.
Ivan had a room full of sharp FileMaker developers for this one, and the conversation carried on well past the session. If you’re exploring FileMaker AI development, I hope you get something out of it too. Reach out with any questions or comments — We are always glad to talk about this work.
Subscribe to get the latest in your inbox.
Built with you in mind
Speak to one of our expert consultants about making sense of your data today. During
this free consultation, we'll address your questions, learn more about your business, and
make some immediate recommendations.
