A thought starter: Xojo, RAD, and AI (towards a possible evolution)

AI
1

Hi all,

Over the past months, several discussions have emerged around Xojo and the use of AI in development, often focused on individual features or specific tools.

I’d like to share a broader thought starter, with the goal of encouraging a constructive discussion among developers rather than proposing a closed or definitive solution.

Xojo is appreciated by many for its RAD model: visual designer, event-driven programming, and rapid development of business applications. This is a real strength and not something that should be lost.

At the same time, the way many of us work with code is changing: external editors, automation tools, headless builds, and AI helping to read and modify complex projects. The question may not be whether Xojo should “chase” these tools, but how it can coexist with them without losing its identity.

With modern AI tools (ChatGPT, Gemini, Claude), I’ve put together a technical document that explores this direction, starting from a few general principles:

  • keeping RAD and the visual designer at the centre
  • avoiding imposed architectures or radical rewrites
  • making projects more open and interoperable
  • moving forward in phases, without disruptive changes

The document is intentionally aspirational but explicitly phased: it’s meant as a basis for discussion, not as a list of features to be implemented immediately.

For those interested in the technical details, I’ve attached the document to this post as a ZIP file.

I hope this can serve as a starting point for a community discussion:

  • Which aspects do you consider most relevant for Xojo’s future?
  • What feels realistic in the short term, and what does not?
  • Which pain points do you experience most in your day-to-day work?

The idea is that value emerges from the discussion itself. Xojo has grown in the past thanks to this kind of exchange within the community.

Thanks to everyone who wants to share their thoughts.

Xojo_RAD_AI_RFC.md.zip (6.7 KB)

1 Like
2

Both the text format and xml work fine with AI. Not sure why you need a new format.

Bidirectional Filesystem Synchronization is necessary.

A CLI would be welcome but IDE Communicator with Json output has been working fine for the last half dozen years or so.

The rationale for both Project Structure Export and Explicit Event Wiring is not clear to me at all.

Just get rid of Jade.

There is only one person you need to convince and that is Geoff. Currently, I would settle for getting the AI code back to Xojo in a faster way.

5 Likes
3

Thanks, Beatrix — I agree with several points.

On formats: you’re right that both the current text format and XML are readable by AI. The reason I mentioned a UIDL-like representation isn’t “because AI can’t parse XML or plain text”, but to make UI state deterministic and diff/patch-friendly (stable IDs, explicit non-default properties, predictable serialization), so external tools and agents can change UI safely without breaking projects.

On bidirectional filesystem sync: fully agree — that’s the key enabler.

On CLI vs IDE Communicator: also agree that IDE Communicator with JSON output can work today. The CLI idea is mainly about a standard, headless, scriptable path (CI/agents, consistent exit codes) without requiring the IDE to be open.

On Project Structure Export: the intent is context reduction. A “structure-only” view (types, modules, method signatures, UI outline) gives agents a map of the project without shipping the whole codebase every time — especially helpful for large or legacy projects and token/cost limits.

On explicit event wiring: for humans the designer works well, but for automated refactors and patches it’s opaque. Making the wiring explicit (or at least exportable) helps tools understand event flow and apply changes more reliably.

On Jade: I’m not attached to it as a chat feature. My focus is on interoperability; Jade could evolve into an optional orchestrator or bridge, or simply be deprioritized if it doesn’t add clear value.

And yes — getting AI-generated code back into Xojo faster is a very reasonable first win.

Thanks for sharing your perspective.

1 Like
4

Looks like he is a big fan of useles features.

If I need my own api key and jade cant interact with the code. It is easier to work directly on the own AI chat. WTF is the use case for jade ?

2 Likes
5

it was explained in this forum. a initial build-in.
i think 2026 there will come more modifications.

6

https://www.youtube.com/watch?v=NA01uEhdjSw

Section talks about Jade

7

link with correct timing to ai

https://youtu.be/NA01uEhdjSw?si=Wyvv173IjRwHSX6R&t=1878

8

Basically Jade is a chat section in a browser window pointed to claude.ai . We already have “Jade” for all previous Xojo versions if we want to.

2 Likes
9

and as you need a claude api key it cost more as it is not included in your eventual claude plan (you still need to have one) use claude free web chat and you have the same as Jade but it costs you nothing.

1 Like
10

That is the first time I read that.

11

of course you need to have a “good” prompt
but Claude knows what is Xojo quite well
only tell it to look into documentation.xojo.com for the exact syntax and it works very well

12

Yes, the Jade functionality is a start.
Maybe it’s a slightly too cautious start, or perhaps it is released a bit too early ;-).
But hey, it’s just a button in the top-right corner—if you don’t click it, it won’t bother you.

Yes, it’s limited. Maybe in 25.3.1 there could be an option to set a maximum number of tokens… Right now it stops halfway :frowning: which results in a lot—a lot—of copying and pasting.

But I see it more as a signal from Xojo that they are moving forward and becoming more mature. Bidirectional text files are especially important, and being able to generate classes from files and folder structures instead of importing or copy-pasting would be SUPER.

So, really… I’m happy. I’d just like to see it on the roadmap (I couldn’t find any of this there yet).

13

I don’t think it is. The problem is it needs a Claude API key (which is pretty expensive, most certainly if you use it a lot).

Claude can be used for free with Cursor (or even the Claude chat will do fine if you do not mind copy/paste code).

I really do not understand why they put resources in Jade. Unless it can be used for free with Claude (or any other AI) it won’t stand a change.

2 Likes
14

Free! That should probably be illegal.

Teachers create assignments with AI, and students complete them with AI.
Managers make company policies they don’t even understand themselves (but hey, AI said it’s correct, so it must be true—Arghhhh!).

All of this relies on AI’s “intelligence,” which, as we should know, is based on the most probable outcome—the “average of the herd” (aka humanity on the internet).

To keep AI running, Meta alone operates 1.3 million GPUs, consuming megawatts of electricity.
And for what? 90% of it is pointless questions.

I suspect OpenAI and Google are even bigger players in the AI game.

Let’s do the math:

  • Average power of a nuclear reactor: 1,200 MW (1,200,000 kW)
    Max power consumption of an NVIDIA H100 GPU: 700 W = 0.7 kW
    GPUs per reactor: 1,200,000 / 0.7 ≈ 1.7 million H100s at full capacity

Sure, I know—network losses, cooling, infrastructure, etc.
Nuclear plants usually operate at 90–95% capacity, so realistically: ~1.5 million GPUs.

And that’s without counting the servers these GPUs run on, powered by Epyc or Intel processors, plus storage and networking.

So basically, one nuclear reactor could be powering just Meta AI (not even counting all the other wasteful activities of Meta #SavingThePlanet :thinking: :face_with_peeking_eye:).

I have nothing against AI, as long as people keep thinking for themselves.
But I do have a problem with “free AI” for entertainment.
How do we expect people to be mindful of resource usage if they don’t have to pay for it?

So..I’m not some leftist world-saver, but it’s ironic how many “green” activists are happily running AI at full throttle.

PS: And I’m writing this while we (my fellow countrymen and I) are portrayed worldwide as cheapskates with the term “Dutch parties” :rofl:

15

Thanks for playing the moral knight for a moment. :roll_eyes:

1 Like
16

Cursor has a monthly subscription with AI limits based on the price tier. It isn’t free but it’s included in the product’s subscription.

17

I use the free tier of Cursor. For most projects it is good enough.

18

Do you need a API key to use your Cursor account???
Yes? problem solved you can use Xojo JADE
No? Oh oh … Houston we have a problem :slight_smile:

19

Oh yes, I was a moral knight for a very brief moment.:innocent:
I used AI to translate my Dutch text. But fair is fair, I paid $50 for Claude, and it works perfectly. A year ago I paid $5 for Claude, and you could ask quite a lot before those $5 were used up.

But I don’t chat like, “Hi Claude, how are you doing today,” and that helps a lot with my mental health, my wallet and the planet.

20

No