I'm struggling with Azure's AI implementation. It feels like a confusing mess with all the different AI products that seem to do similar things. Whenever I switch models, I often get a 'resource not found' error because the URL they provide doesn't match their code examples. I've searched everywhere for the correct URL. Even when I ask ChatGPT to help me with working code using their documentation link, I still get stuck. I'm baffled by the version dates too. Why is it so complicated when the only required parameters should be the model name, URL, and API key? Additionally, I run into issues when I try to train the model with inaccurate data. I had to revert back to my local Ollama server just to have things function properly again.
6 Answers
Without a doubt, it’s a mess. The best way to approach this is to use Azure Foundry just as a model provider and pair it with the best tools out there. For instance, the Claude Agents SDK is great for abstractions, and combining various vector stores can help with memory setups. Just don’t let cloud providers dictate your tech stack or you’re in for a rough ride.
Microslop a mess? Say it isn’t so! Seriously though, these issues are getting ridiculous.
As someone without formal Azure training but 10 years of dev experience, I just wish I could get the basic functions to work smoothly.
Yeah, it’s definitely a bit chaotic. It has its bugs, but you can make things work with some tweaks here and there.
I feel you! It’s frustrating when you put in the effort to get things running and you hit a wall. I'm a product manager working on documentation, and I'm all ears for your feedback. I want to ensure we fix this mess and make it easier for users like you!
Are you guys hiring? I got laid off last fiscal year and would love to help fix these issues!
Honestly, it feels like there's hardly any one on the front-end to help out. It’s ridiculous!
We've been working on a next-gen Agent Builder portal, so check it out at ai.azure.com/nextgen. We’d love to hear your thoughts!
Definitely fix the Foundry agent VNet injection issues! It’s currently only deployable with Bicep, and even then, it fails a lot. The documentation is lacking too, which makes it harder for companies like ours to effectively use Azure. We’ve been considering alternatives because of these challenges.