Speed Up Your AI Development Workflow by 2x

Hello everybody, I'm Nick here. In this video, I want to talk about how I double my productivity of my development workflow by utilizing a couple of applications that really speed up what I think is my bottleneck, and I'm sure it's the bottleneck for many of you, and it's mainly related to typing. Now, because we no longer go into an application and we just start typing everything manually, like you'll still do that, but a lot of that has now been dominated by AI, I want to see how I can improve my words per minute. And I can type fairly fast, uh not super fast. I can do 100 words per minute pretty consistently, but in code with auto complete and so on, things slow you down more than you think. It's not like you're typing down an essay. So, that word per minute is actually reduced quite a bit. Now, going back into an AI agent like Claude, for example, you wouldn't write code here. You might paste code, but you would write what you want the tool to do. So, for that reason, I looked into solutions to how I can solve it. And I found two things. First, sometimes I just want to type down my thoughts. And I can go ahead and say in this demo, which is part of the demo for my online workshop I'm running in 7 days. If you want to join, there's a few links in the description. You can check that out. I'm going to teach you how to do vibe coding for production at proper scale, something you can actually push to production to customers and be secure and safe about it. And as part of that workshop, I'm showing, "Hey, here's an application uh and ignore the text being small. The code doesn't really matter, but this application has a few vulnerabilities I want to showcase, and I want to show how AI can sort of fix that by auditing it and then suggesting what we should do about it. " Now, I can go here and manually type, "Please go ahead and run a security audit for my code. " However, can I speed that up? And the answer is yes, and I can do it in two ways. The first one is with auto complete. And not stupid auto complete that tries to hallucinate things based on no context or minimal context, but actually take a look at what's going on in the screen and what I'm typing, process that, and then start suggesting things based on that. So, for example, if I now start typing, "Please run," you can actually see text is appearing. It's small because the scaling doesn't work well in this big screen. When it is smaller, it works fine. But if you notice here, it says, "Please run the following command. " It's suggesting. And if I was to press tab, it will go ahead and accept this word as something I typed. But I want to say, "Please run a And it suggests code analysis. If I say "S," it's going to say "Security check on the code. " And I can press tab tab tab and it says below. I don't want to do it in the code below. I of this code base. So, "Please run a security check on the code of this project. " And I can just accept it, and it auto completes the words using the context. Now, how is that happening? It must be some LLM involved. And the answer is yes, there is. However, I'm not the biggest fan of LLMs seeing what I do on my screen and then sending it into some server for processing. So, all of this is actually running locally. And I'm achieving that with an application called Code Typist. So, Code Typist is this app over here, which I'm sorry it is small. I'm going to zoom in a bit. But Code Typist is an app I'm running on my Mac, which, as you can see, is using Gemma 4E4B, which is a local LLM. At no point is a request sent to the server for processing. All of this happening locally on my machine, and I can choose how much context I want to give it. I can give it some information, so I can personalize it. For example, I've said a few things about me. For example, I'm Nick Chapsas. I usually write in English. Please write in a friendly and casual tone. And so on. And you can customize the custom AI instructions. And then you can choose tons of things. For example, don't use emojis on whatever you're writing because they're cringe. Or how much context. Hey, use screenshots for context. Do not use the clipboard for context. And it will take screenshots, and as I'm typing, it will look at the context of my screen and then do all that processing offline. At no point does it go in OpenAI or Google or any other service, which I feel very comfortable about because it's all local here, and then it will use that to suggest things on everything. It's not just for coding. If I write an email, this will kick in. This will basically kick in everywhere. When I'm on Notion and working on some notes, it will do it. When I'm writing a message on WhatsApp, it will do it. And again, all of this is local. And you can choose the model, so you don't have to use Gemma 4. It gives you a few suggestions on other models that are working. Depending on

how good your machine is and how much RAM you have, you can choose to have a bigger model. Of course, a bigger model means that it might be slower, but my machine is fine. I have half a terabyte of RAM, so I'm way more than comfortable with using something like this and not stressing the system. And I have found that this significantly speeds up not just my coding workflow, but sort of any other workflow I have where I need to type because the context is so good. It understands what's going on. And this doesn't just work at the terminal, it can work anywhere. I mean, you can limit it. So, you can go all the way here to app settings and say, "Don't work on Rider. Don't work on IntelliJ. " because those things have their own auto complete usually. So, you can exclude a few things that you don't want the thing to kick into. And you can also do the same for domains. So, if you want to exclude a specific domain, maybe your banking domain and so on, it will know and it won't go ahead and do anything there. Now again, this is still fine because all of this is local. How do I know it's local? Well, first, the model is local, but to make sure that it absolutely doesn't send anything over the wire, I also have Radio Silence installed, which allows me to block certain applications from ever calling the web. So, for example, as you can see here, Code Typist is blocked. It can't go over the web and send requests. You can actually go to network monitor and guarantee that nothing will go to the cloud on any other server. So, this is the first part. The second part is that, "Look, I'm working in an office alone. Why do I have to type all the time? " And the answer is I don't. Usually, now what I do is I dictate. So, I talk to my microphone, and in fact, I have a small microphone over here, which I always have on my desk separate from this because I don't want to have this in front of me all the time. And this can now pick up exactly what I'm saying by me just pressing caps lock, holding it down, and space. I'm using Hyper Key, which is a utility on Macs to do that. It gives you a more flexible way to do key bindings. And if I go and click it, you'll see a window pop up at the bottom, and it will start writing down what I'm saying. So, just like that. And if I just delete it, if I show you how the workflow would be, and you saw how fast it is, then I can simply say, "Run a security audit for this application. " That's it. And it's there. This significantly speeds up my workflow. Of course, I can't talk every time. If it's late, I don't want to be whispering, "Please run a security workflow for this application. " and so on. But when I'm here and it's the day, I want to do that, and it's here, and it's super fast. And in the same nature as Code Typist, this is using an application called Type Whisper. Now, with Type Whisper, again, this is a local thing that, as you could see before in Radio Silence, it cannot actually access the web. Everything is here local and secure. And if I go back, I accidentally closed it. you'll see that I have chosen over here a default engine. I'm using ParaKit cuz I found this is the fastest and most accurate engine for this specific solution I have, but WhisperKit is pretty good as well. It's just a bit slower. And then you can choose the model. Those are local models, which you can go over here on integrations and install. So, I have WhisperKit installed. I have Apple ParaKit Speech as well, which is extremely fast, but not that accurate and contextually intelligent. And I can also have local file memory, so it can remember a few of the things I've said to use them as context. I can also have snippets. For example, when I want a hyphen to be converted into something else. And I can go here and say, "Nick hyphen Chapsas. " And it will chaps. And it will go here and replace hyphen with that instead of actually typing the word hyphen. I'm doing the same with things like front-end design, which is a skill you would use in Claude or Cursor. And so on. You can have your own dictionary. You can customize everything about this. You can have file transcripts. You can have uh custom hot keys. As you can see, I'm using this very elaborate one, but these four ones are actually just holding down the caps lock button. And you can customize everything about it. And again, this is app-aware, so it can actually understand, "Hey, I'm in an editor. I'm in a terminal. I'm somewhere. " And it will adapt what I'm writing to whatever it can hear. So, if I want to be a bit more elaborate, I can say, very quickly, and that's impressive about it. It's actually very snappy. I can say, "Please run a security audit about this application. Make sure that SQL injection is not possible with this application, and also check for any HTTP-related vulnerabilities that might be affecting an application like this. " So, I'm speaking in my very normal voice, and it's typing way faster than I could possibly write this. If I try to write this, I mean, it's not going to

Okay, this is actually reading the context and it's going ahead and it's typing it with Code Typer, but as you can see, it still isn't as fast as the local LLM using my voice. It's lovely. It is amazing. I know tons of you are on Macs, and I'm certain there's alternatives for this on Windows, but I'm not using Windows for development, so I can't possibly know. But, as you can see, this can significantly speed up your workflow. I'm hitting 150 to 160 words per minute when I'm dictating with my normal voice, and this does not have to go to the cloud, involve that latency, and involve your recording going somewhere, which I think is lovely. But, now I want to know from you. Do you use any tools like this to speed up your workflow? Leave a comment down below, let me know. Well, that's all I had for you for this video. Thank you very much for watching. As always, keep coding. Dictating.