A New RAD Studio Workflow Is Coming

,

Really?

I have a short story, years ago, well before AI, we were upgrading our web site, to something with all the latest ‘stuff’.

I wanted to be able to interface to it from Delphi but had no idea where to begin, so we added ‘a Delphi Sample’ to the list of requirements.

The contractor said, “I used to code in Delphi, but haven’t for 15-20 years, I’m not sure if I could create a Delphi sample, but I can try”, he quoted a week to do it, but warned it might take a fortnight. We okayed this.

It took him one day.

I think the move away from Delphi caused him to learn something new. Something that could have been applied to Delphi just as easily if he’d continued his Delphi learning, rather than maintaining the stasis his coding had been in for years.

This may be old news by now but Kai is Chucki. For those who weren’t aware of Chucki, which included myself until a few hours ago, Chucki was announced (though I’m not sure if it was released to the public or not) in January this year.

It was written a name many would be at least somewhat familiar with, Roman Kassebaum. Is Kassebaum the “K” in Kai perhaps?

The Chucki website looks to have been taken down but there is a demo video on YouTube from back in January. I don’t think there’s anything in that demo that wasn’t shown in yesterday’s webinar but it’s nice to see the creator demoing it.

I don’t know if Roman is continuing to work on it for Embarcadero or if he has handed it over to them and walked away.

As for why there’s an additional cost I agree it’s not ideal. The most obvious reason is probably the actual reason but if I was being extra generous to Embarcadero, having it funded through a separate purchase means they can say it’s not taking resources away from the main product.

For an AI hesitant customer base, like the Delph community appears to be, that could carry some weight.

Oh wow. Yes, they released to the public (well I didn’t purchase but they had pricing, purchase links, etc). I had mentioned it here in the past but it wasn’t a widely known product, largely because it was fairly new I guess.

The Chucki website was still up until recently - a bit strange that it is just gone without a message or even redirect.

When I looked at it I was reasonably impressed with what they achieved in the RAD Studio IDE. The editor can be quite a challenge to extend (I have ideas that I sadly don’t think are possible due to the limited API) and I had wondered about how it makes changes to opened forms and code behind the IDE’s back and getting the IDE to nicely refresh or not show errors if the form temporarily contains partial changes/unfixed errors.

There are still gaps when compared to a more mature/feature rich tool such as Windsurf but this is a great step and hopefully will provide a robust and helpful experience and grow.

Yes, Kai did grow from Chucki. Roman likes to say that the K in Kai stands for Kassebaum but that’s a happy convergence. We chose the name, as we do for all the betas and release names, from a list which gets thoroughly vetted for ‘unfortunate’ meanings in the various languages into which RAD Studio is translated. We did have some others, and one in particular which nearly ended up becoming final (which I was not a fan of, but luckily was voted down by enough of us that it never saw the light of day).

We have secured a long-term deal with “Team Kassebaum” as I like to call Roman and his sons. We have other team members who work on it too - but Team Kassebaum is completely dedicated to Kai, and only Kai (they do not work on RAD Studio etc). Kai also has its own PM. This is one of the reasons why it’s a separate subscription. I absolutely wish it was included in the price, but, right now, we had to charge for it in order to ensure we could get internal funding for it and also make it worthwhile for the Kassebaums to commit resources to it at the level necessary to make it viable, and to evolve.

It is not quite the same as Chucki, especially behind the scenes, and as it evolves to meet changes in the AI providers and how people use AI it will step very far into a new future for it.

We had many people come to us with AI solutions of all sorts of types. There are lots of credible projects out there - for example David Millington’s CodeBot, is awesome. We believe that we should compete by offering what we (Embarcadero) think will work for our typical customer profile - and also to meet the suggestions given to us by various universities and so on.

We spent many months talking to people and listening to criticisms, suggestions, pleadings, rants and raves. Will Kai be what everyone wants? No, absolutely not, and there are plenty of people like @hsvandrew where we may never be the solution to anything (even if I offered him a free license for RAD Studio and Kai) - and I totally get that. But, based on the research we put in Kai is going to be the answer to a very great deal of other people’s wishes and suggestions.

Personally, I think the cloud providers like Anthropic, Google, et al are going to continue to fight it out - and, hopefully that competition will benefit the developer community overall.

What you see in Kai right now is very unlikely to be the same in a very short space of time and this will continue to be the case. It’s inevitable, given the volatility of the development industry. The same is true for RAD Studio, it’s going to continue to adapt to what developers want and how the development industry is morphing - but the settled reality of that; we don’t want to park our trucks on wet concrete. Number one is quality. Number two is the reintroduction of features which were deprecated. Number three is everything else.

I love Delphi. Always have from the very first day I saw it. Nowadays I get to help shape its future. And, yes, AI will help shape that future too.

@ianbarker You are correct. A free license wouldn’t help. I did pay for a pro or enterprise licenses for Delphi though since I was 15 years old until my current one expires, and that will be it.

I also spent more yesterday on Cursor tokens than an entire yearly subscription to Kai.

My point was about you coming to market a year after everyone moved to fresh delicious wood fired pizza at $20/pizza (or pie as you might call it) which includes tokens, and you offered up stale frozen pizza for I think about $250USD/year without any tokens, with a UI from 2010 and really seem to be underestimating how much work has gone into IDE’s over the last 18 months.

Just to even catch up now would take a substantial investment and speed to market that’s never been witnessed before, at a time when the product at least in terms of jobs in English speaking, western markets, is dead and I’ve seen no signs of any major western companies building anything substantial and new in the product for years.

Considering Cursor’s annualized revenue is around US$2.7B, and Embarcadero has achieved very little in 18 years, I won’t be holding my breath :).

With the only green shot I could see (David Millington) not having stayed, that tells me more than any marketing spiel can pretend is going on and the job numbers don’t lie like the filtered marketing campaigns that block all comments unless they are positive pretend reality to be. Isn’t it funny how your YouTube videos have not a single question/comment.

Alister Christie did a first impressions video of Kai that I felt was a much better demo than what was shown for the webinar. I felt like that demo showed how people were using AI 12 months ago.

I think AI is both a risk and an opportunity for Delphi/Embarcadero. There is a risk that AI could be used to move “legacy” Delphi off onto “modern” languages like C#, Javascript, etc. It could just as easily be used to bring those legacy Delphi apps up to the current supported version of Delphi. @ianbarker you could do a webinar about using AI to update legacy code to the current version of Delphi.

I had a good play with Kai (in Māori it means food, I’ve been told not to play with my food before…) over the weekend, using Copilot. I didn’t get any interactions with the IDE, it updated units and forms, but I had to close and reopen the units to get the changes, also no auto compiling to test like I’ve seen in the demos. Is this just a CoPilot issue?

Code completion/suggestion was very good.

Tried with Ollama locally, but way too slow to be useful, and I’m not forking out $5k for a video card just for that…

Regards,

Yes, it’s on the list. Thanks Geoff.

Yes, we chose Kai because it means quite a fair few positive things in various languages. My favorite is the Welsh meaning “keeper of keys” - pretty appropriate really, lol. My Welsh grandfather would get a kick out of it.

I’m not sure about this. I am not aware of a specific problem with CoPilot. I’ve been using Claude, Gemini, Ollama, and LM Studio, all of which seem to work (the local models needed a bit of tweaking, especially with the context window size). I showed them working at a user event in Portugal earlier today.

I am doing a webinar next week about configuring various providers - and especially local ones. I suggest you try using GitHub - AlexsJones/llmfit: Hundreds of models & providers. One command to find what runs on your hardware. · GitHub and see what models it suggests. I also have another tool which was shown to me today by the developer. It’s being worked on but should be ready next week too and might solve quite a few problems.

This is release 1.0 of Kai. There has been a tsunami of various comments (some against, many for too). The Restream system doiked all the YouTube comments into a moderation bin - we’re not quite sure why - but we’ve fixed that too and pushed them back out into the daylight, good or bad.

Is this just a CoPilot issue?

All of those things require tool calling, so I suspect that CoPilot doesn’t support that (not all LLMs do). Kai would register a lot of tool functions that let the LLM decide when it wants to compile, what ToolsAPI functions to use to interact with the IDE.

Makes sense…