an for the history
erci schmitt, ex ceo google, said recently, that steve jobs was like a a diva in recent interview
i hear a lot of people saying developpers behave like divas sometimes
last on on steve jobs, is kein leary saying steve jobs “not a nice guy”, well…
What you’re doing is called “goalpost moving”. I have no idea, nor interest, in what you are trying to prove at this point…
to be technically correct on the side of semantic, i asked chat gpt
Here’s a quick summary of what happened:
So, while he wasn’t directly fired with a pink slip, he was effectively pushed out of Apple in 1985.
AI for the average person and AI for a developer using a development tool are very different things. Apple needs to and will almost certain do much better than their initial AI offering. They appear to be keenly aware that what they have delivered isn’t nearly good enough. In fact, it’s poor enough that my wife turned it off entirely. The success of Apple is in part how tenacious they are. They will keep working at things they know are important and at which they are confident they will eventually succeed. AI is one of those things.
I agree with your assessment that for a developer, AI should be there to assist when needed much in the same way that another developer working shoulder to shoulder with you on a project would. When you can’t remember the name of a particular class, or how to use the 3rd parameter of a method or the max value of a 32 bit integer or where the heck is that +1 bug in this code or how to get started writing a method that parses a particular type of data file, that’s when you want to be able to choose to ask an AI Assistant for help. And when it does help, that help should be clear, obvious and instantly undoable if it turns out to be wrong. Our approach to AI will no doubt take several releases to complete. It will be an incremental approach. There should be no doubt in anyone’s mind that our world is in the process of changing because of AI. I would hesitate to try to predict where it will go in 10 or 20 years but it’s useful enough as a software development aid that I believe it to be a very high priority for us to integrate thoughtfully into Xojo.
I’ve experimented quite successfully with Claude Sonnet 4 and Xojo. The key is to integrate AI in just the right way so that it’s truly useful for the developer and not just some useless branding feature.
This is exactly right. The public LLMs are very good. They weren’t so good at Xojo initially but they have gotten a LOT better especially with some good prompting. There are also ways we are looking into to help the LLMs learn Xojo even better than they know it now.
This drives me crazy about AI. An assistant that is occasionally, confidently, subtly wrong and requires me to carefully screen its work isn’t an assistant I have much interest in using.
I’m probably in the minority here with my old fashioned notions of the value of human intellect and creativity. But, then, I’m not running a sweatshop-style code mill like Amazon.
To be accurate in my post, I used my eyes and brain and read an article to refresh my memory, and sure enough, you were wrong to state he was fired. Then I used what I just read to synthesize a post explaining it in a way you could understand.
But I guess I’m probably just old and afraid of AI in your book. C’est la vie, as they say.
Thank you, Geoff - I completely agree with your overall vision. But if I had to pick one single aspect of AI integration that I believe is by far the most important, it would be code autocompletion - and I don’t mean just suggesting method names or parameters.
I mean the kind of inline, context-aware suggestions that tools like GitHub Copilot already provide - where I type the name of a function like ParseInvoiceCSV, and Copilot instantly suggests a full implementation based on the project context. That is 90% of where the productivity gain comes from in everyday development with AI - not through asking questions or writing prompts, but through deeply integrated, predictive code suggestions as I type.
Everything else is secondary. Without this kind of autocomplete support, most of the power of AI for developers is simply lost.
Honestly, I face the same issue with human coworkers all the time - confidently wrong, and I still need to review everything to avoid problems. The real question is always: Am I faster doing it myself, or reviewing someone else’s work?
With AI, at least for me, there’s no onboarding phase - and from day one, it saves me about 70% of my time. That’s a trade-off I’m very happy to make.
I agree that it can be very frustrating when AI gets it wrong. In the beginning and honestly up until quite recently, I thought it was wrong enough that it was just not ready for prime time. My opinion on this has changed. I’m not saying it will be perfect when we make it available in Xojo but it’s going to be useful and it will get better and better over time. At first it might be like having a junior software developer that can help you get grunt work done more quickly. But overtime, it will reach the point where it will be like having a seasoned, experienced developer at your beck and call for those moments when you need some help.
We will get to having it help us improve autocomplete. That’s not where we are starting but it is part of our vision for it.
I used to have the same attitude about the self-driving feature of my car. Monitoring it all the time in case it made a mistake or drove annoyingly (which it did a lot) was more stressful than just driving myself (which I actually enjoy). But the past 12 months of advancement have changed my mind, and I now use it almost daily - it’s nearly perfect.
An AI coding assistant can be very useful even if you don’t want it to generate code - you can use it to find bugs in your code. Not long ago I was banging my head on for an hour or two on a routine that didn’t produce the expected result. Finally I uploaded a screenshot of the IDE to Claude and it instantly saw that I’d misplaced a decimal in a constant.
But you chose to go to Claude and use it, chose a self driving vehicle and chose to continue experimenting with it. Which is awesome, I’m jealous of your car.
My hang up isn’t on the technology, it’s on the forced usage and integration. The terms of service with the what’s yours is mine and what’s mine is mine clauses.
The always online monitoring and subscriptions that are optional but not really.
If the world in general can just chill and treat me like a customer again I’d be totally happy with a lot of things. I miss the tech world of the 90s early 2000s lol
Indeed. At my business, we pay for Google Workspace (Gmail+Google Docs etc) and this month the monthly price went up because they are foisting their Gemini AI ■■■■■■■■ on all their customers. I have no interest in this functionality, but it’s a bundle and there’s no way to get rid of it.
I fully agree. I really don’t need Adobe Reader constantly nagging me to use AI to read PDFs and Outlook pestering me to let it write my emails for me 
Nearly perfect is the difference between your life and your death. I have worked 20 years in the automotive industry and I’m not going to trust any self-driving car in this lifetime.
It’s not like I’m getting in the back seat or going to sleep (yet), I’m still “supervising”. But it just drove me door-to-door 140 miles each way to my sister’s house and back, on highways and back roads, flawlessly. It’s an amazing achievement that has to be experienced to believed.
I’m a radiologist and use various AI tools a lot in my daily practice and am involved in the implementation of some cutting edge stuff in my day job. I also use LLMs for coding with Xojo frequently and, whilst they are far from perfect, they have long since crossed the threshold into being indispensable. There are limitations with the Xojo knowledge of the frontier LLMs but even so, they are helpful.
I think Xojo Inc needs to iterate quickly on this which, unfortunately, has not been their strength historically. There needs to be a way to expose code in the IDE to a LLM more easily. Currently I have to copy and paste stuff into a LLM vendor’s website or Open WebUI and the LLM has little to no context of the rest of my project. We probably need new IDE scripting features that we can expose to MCP servers.
My fear is that Xojo will take too long to release any form of integration. I absolutely think it should be a feature that can be ignored by some users. I don’t think Xojo Inc should be bundling its own LLM variant. I feel they should allow the end user to plug in their own API key (for Anthropic, OpenAI or even Ollama). Heck, I’ve even released Xojo code they can use if they want (GitHub - gkjpettet/AIKit: A Xojo module for interacting with AI tools.).
I’ve been using ChatGPT as a pair programmer. It has worked quite well. As long as one understands the current limitations of a deep expert system, monitors context tokens and keeps the conversation well focused, I’ve found that it accelerates my coding proficiency. Clearly, with less information from which to train, working in Xojo is not as effective as, say, writing C# (and there’s also copilot, but that’s different from a pair programming conversation as opposed to filling in the blanks), but still, I can tell ChatGPT that I’m in Xojo on a Mac, and get reasonable suggestions and improvements to my codebase.
While copilot is nice, for finishing sentences, so to speak, sometimes it is annoying and one has to be careful not to hit a key at an inopportune time, causing a block of code to be inserted in an inappropriate place. I’m interested to see if ChatGPT improves its Xojo expertise if more people use it like I do. Should an expert system be part of the Xojo IDE? For me, that’s an open question. It’s quite an investment of development time to include it and keep it updated. Will it be a subscription service since there’s compute time behind it, or will Xojo host a proprietary server? We are at an interesting crossroad.