My time with Claude Code and what comes next

What's up, everybody? Welcome to another episode of the P W podcast. In this one, I'm gonna talk about AI, but don't worry. I'm not gonna talk about it and go crazy like all the other places on the Internet. I have been using Claude code AI for the past couple of weeks.

I went ahead and signed up for using it for month, is $20 a month because I figured, you know, at some point, I'm just gonna have to pay the money and try this thing. And I was very pleasantly surprised. I'll be honest going into it, I thought it would do well but I really wasn't sure, you know, like how much am I gonna use this and is it worth the money, how accurate is it gonna be, how many tokens do you get and all these kind of things. So let's dive into this. The intention here was I signed up for a month.

I am coding a lot more than I normally would in the average month to really put this thing to its paces and I'm also using it to do things that I already know how to do and that is key to my testing here because sure for me using AI is a great way to learn new things and solve problems when I get stuck and learn to do all these techniques. In this case I was using it for a lot of c sharp, for game programming and for cleaning up an astro based website which is sort of a lot of JavaScript for want of a better way of putting it and I knew how to do a lot of these things but there's always those parts that you just can't figure out or you think you know I'm not sure the best way to do this and for me that's where AI comes into play to help and assist me and get things done quicker and learning the process. Well interestingly, let's start with the game develop. It struggled but it struggled with things that I guess it's fair to say are reasonable. Meaning that I was doing some c sharp in unity and I was using a third party library.

Now with that it had all the source code. It had access to all the source code. It even tried to use the source code. It knew how to do it and everything else. But it still couldn't solve my problem with extremely large numbers and in gaming that means things like you know quintillion, trillion, even billion, you know, whatever it may be.

Really large numbers that you wanna mess around within a game, right, an idle type game, those kind of things. It suffered there but it did a great job at doing a lot of fundamental game programming things like timers and dealing with content on the screen. Things that I already knew how to do but hadn't done in a long time and I figured there's probably a better way than the way I know how to do it. It excelled excelled at that and did a great job. Now that doesn't mean it got it right first time but I was able to have a conversation with it in the terminal because Claude code primarily runs in the terminal but not exclusively.

But I was using a terminal along with my text or code editing application whichever way you want to call it and we got through it and and it worked. Now on the flip side, I also used it for my ASTRO websites. Now ASTRO is one of those static site generators. The details for this really don't matter, but there's a lot of JavaScript, a lot of React, and all those kind of very common elements under the hood. It did a great job on this.

What I wanted it to do, I took three websites. I took peterwidom.com, compileswhip.com and my game site project hack.net. And I said to it, you know, all these sites were working just fine before I started. But I asked it go through, find any errors, find any linting errors in particular which is kind of a code formatting thing for those who who don't know and fix them up. Now it took a while obviously, you know, there's a lot of files there especially if you think about a few years worth of blog posts on peterwyden.com and it found a whole bunch of warnings and some errors and it fixed them all.

It it understood it. It realized, okay, I know this is a blogging website. I got some product pages in there. I see you've got a bunch of markdown files to supply the content and it was astounding to me all the little things it found that I had just not seen. Things like incorrect tags and not closing HTML tags.

All of those really silly simple things that modern browsers can deal with. It's still best not to have them there so that everything renders nicely. They did a great job and it also fixed up all the linting errors so things like extremely long lines of code, it broke down into a couple of lines, form formatted all the code nicely, indentation, all these kind of things. And at the end of it, the three websites, were working for the most part. There was a couple of things I had to work through on the compileswift.com website, but it understood it eventually.

And the key to this is really having a conversation with these tools as if you were talking to a person, which I get it sounds weird. Right? But instead of just giving it instructions and expecting them to execute them and saying, okay, this is wrong. I need you to do this again. You really wanna have a conversation with it because the more you give it and the more conversational the flow goes, the better it learns and understands and tries to fix the errors and tries to find better ways to do things.

A review at the end of this is if you do a lot of coding and I say coding because this is Claude code not just Claude as in the AI service. Then I think you're going to find it useful. It works really well in a terminal and if you got a terminal in your code editor of choice which let's be honest is most of them these days, it's smart enough when you run it to realize hey you're using whatever, right, Visual Studio Code, JetBrains tools, whatever it is and say hey I'm gonna work with that for you. So what I mean by that is if I select code in the editor window I can see down in the terminal running in the editor that it recognizes I've selected three lines of code or whatever. So that when I'm having a conversation with the service it knows what I'm looking at, it knows what I've got to open, it's working with that but it's also analyzing and learning from the entire code base in the project as well or folder, whichever way you want to look at it, but think of it in project terms.

So I'm gonna say it's a win. Now, I think it's worth the $20 a month. It's worth noting that there is like a year's worth subscription you can get and gives you more credits and everything else. I didn't hit the limit of the 20 in a month. And I think honestly, you know, for the sake of saving whatever it is, $30 a year to get a subscription plan because I think it's $17 a month if you do every year.

I don't think it's necessarily worth the difference right now for me because there will be months where I don't code a lot or I just don't need it. But I also recognize that this is a very useful tool for me as kind of a pair programming technique and like I say kind of someone watching over my shoulder able to analyze my programs and the code and and suggest things and I can even just say to it, you know, document this code and it does a great job. So I'm gonna keep using it I think for a while and see how it goes and so I'll certainly report back in the future. But for anyone listening to this, if you are curious, I would say give any kind of AI tool, service, whatever. If they offer a monthly plan, try it.

See if it works for you. See if you can make it work with your work workflows or adapt them. Because for me, it's been a great time saver for all the sort of fundamental things that is like, okay, I know how to do this, but you can do it in a fraction of the time that I can type in at the code. So you know what AI, just go and do it and I'll check the code afterwards and make sure that it's safe and what I want it to be. Love to hear your thoughts on this.

And if you wanna come on the show, talk about it, reach out to me, peter witham dot com forward slash contact. Other than that, folks, I will talk to you in the next episode.