There's been buzz around Nano Banana (a new and cool image model from Google) and I wanted to try it out I came up with an idea to create a micro SaaS app for generating images with Nano Banana and accepting payments for the generation with Solana I recorded the first steps of the process, you can take look at that in the video I also tried Grok Code Fast for the first time while building it. Very fast, very exciting model that I want to use a lot more

Architecture in software is hard to even define but here are some popular definitions "Parts of the software that are hard to change later" "The decisions you wish you could get right early in a project" "Architecture is about the important stuff. Whatever that is" Clearly, architecture is important, and it is even more important the bigger your system is Nailing the architecture is hard but AI can definitely help us with that One of the good things is that with AI, fewer things are "architecture" since things are way easier to change now. We can never nail architecture completely but we can still nail some of it. Some of the architecture we can identify early with experience or analysis. Good news is that LLMs help with both. They are great for the analysis part and have no problems with going through a lot of text and specification. They also have a lot of context from a ton of systems and have an idea already on what kind of features are hard to change later. So having a chat with AI when starting a project is very valuable for identifying architecturally important decisions and structures. Then some of that we just can't know beforehand. So for those cases, just standard software engineering practices (that the AI knows quite well) are very valuable. For example, encapsulation, single-responsibility principle, separation of concerns, and so on It's very useful if you can identify clear cutoff points and kinda lines around different parts of the system and then make those separate, independent parts. Systems thinking also helps with this a lot. I have some other posts on that that you can check out. I have more thoughts on the topicc but the post is already pretty long. Let me know if this is interesting/useful and you'd like to hear more

Please watch what Cursor does to your computer! One of the cool features is that we can give Cursor the ability to automatically run shell commands. This is extremely handy and speeds up the development. But this also has a downside: Cursor can potentially run anything in the terminal, including messing up your system, not just with your code. Cursor has "Denylist" feature that theoretically should prevent it from running specific commands. But in practice, this just does not work. Cursor has multiple ways to get around the restrictions and destroy stuff if it wanted to. And to make matters worse, the AI models are so eager to please that if given a bad prompt, they will just create a shell script or something instead of slowing down and asking the user for a permission. This becomes (potentially) even worse with prompt injection attacks (Cursor reads something that makes it to do bad stuff) combined with unrestricted access to the internet. I have no solution for this right now. Just please check what Cursor does on your computer. Background Agent doesn't have this problem as much as it runs in a separate environment that can be destroyed. MCP seems promising for fixing this but it is still early and has a bunch of issues that need to be fixed first. You can read a more in-depth explanation about this here

Cursor just released 1.3 Here are some highlights of the new release: - View context usage in Chat - Agent now shares the native terminal with you The update also introduces some smaller improvements, such as faster edits, checkpoints for notebooks, and more Check out the full changelog here

https://youtu.be/A5r2tfifYfY?si=0JDv9IlB-hn-iwqR Is this true? @Toivo Mattila or is it just an outdated video? Thoughts