Crop diseases are honestly a big issue right now, especially when we talk about food security worldwide. A large portion of crops—something close to 40 percentage gets lost every year even before harvesting. This mostly happens because of pests and different plant diseases[11]. In most cases, farmers still rely on traditional methods to figure out what’s wrong with their crops[1]. They look at the leaves, the color, or sometimes just go by experience. Some even consult experts if available. But the problem is, this whole process is slow and not always accurate. It really depends on who is checking. And in many rural areas, getting expert help is not that easy either. Things have started to change a bit with the use of technology. Artificial Intelligence (AI) and Deep Learning (DL) are now slowly entering the agriculture field[13]. Instead of manually checking everything, these systems can analyze crop images and detect diseases automatically. It’s faster, and in many cases, more consistent too. That’s why people are now talking more about precision agriculture—it’s becoming more practical than before. In this study, we went through several research papers related to AI-based crop disease detection. Most of them are using CNN models (Convolutional Neural Networks)[13], which are quite common in image-related tasks. Apart from that, models like ResNet, EfficientNet, and MobileNet[14] are also being used quite a lot. Recently, some researchers have also started experimenting with Vision Transformers (ViT), although this area is still developing. Another thing worth mentioning is that many of these systems don’t just detect diseases anymore. They also try to suggest possible treatments. Different approaches are used for this—some follow rule-based methods, while others use filtering techniques. A few studies have even used models like GPT-3.5 to generate recommendations[5], which is interesting but still needs careful validation. If we look at the results, they are actually quite promising. For example, EfficientNetB5-based models have reported accuracy above 96 percentage[6]. Similarly, VGG-13 models are giving around 95 percentage accuracy[9]. But still, there are some clear limitations. A lot of these models are trained on specific datasets, which may not fully represent real farm conditions. Also, testing across different environments is still limited. On top of that, deploying these systems in real- world farming situations is not always straightforward. There are practical challenges involved. And one more important point—whatever suggestions these AI systems give should still be verified by agricultural experts before farmers fully rely on them. Overall, this field definitely has potential, no doubt about that. But it’s still evolving. Going forward, it would be useful to focus on making these systems more understandable (like using Explainable AI), using drones for monitoring large areas, and building applications in local languages so farmers can actually use them easily. Better evaluation methods are also needed soresults can be compared properly across studies