Like many of you, I suspect, I've been using AI tools to assist me with tasks both mundane ("Help me find a cat bed large enough for three cats") and complex ("Configuration recommendations for pfSense"). There are obviously tons of choices out there for AI tools, but I chose Claude—I pay for the least-expensive plan, the Pro level at $204 per year.
Until recently, all my interactions with Claude had been through either the web site or the macOS app. I was aware of Claude's Terminal-based programming-focused tool, Claude Code (and its corresponding tab in the Mac app), but hadn't tried using it yet. Why not? Because I'm not a programmer, so I figured what could it do for me?
But in the last few weeks, I've thrown some projects at Claude Code, to see what it could do. In the end, Claude Code created things that I might have been able to make on my own could never have created with this level of usefulness and completeness. And Claude Code did it all in a matter of hours (spread over some number of days), while I simply told it what I wanted, and then how to refine what it had delivered.
So what has Claude Code done for me? I used it to bring my macOS Release Dates and Rates post back to life, in a version that's incredibly easy to maintain. I used it to create a simple site monitoring script that lets me know when I need to update the macOS releases post.
Those projects were, I thought, impressive. But they were nothing compared to these two…
(I'll post more details on both projects, including download links, in soon-to-arrive posts. And, actually, those two are now trivial compared to my latest project, but details on that one will remain unshared for now…)
The Queue (left) is a database-driven tracker for TV shows and movies that I'm watching or want to watch. Claude Code created this—from scratch, with just the prompt I provided it—in about seven minutes, and it worked perfectly the first time. I then spent a few hours refining what it presented, adding features, and making it work just like I wanted it to.
Dramatis, on the right, collates main character info for books, movies, and TV series, and stores it locally in a database. With Dramatis, I can now easily see relationships between characters, characters grouped into buckets that make sense for the show or book's plot, and (optionally) reveal spoilers for certain characters. This makes it much easier for me to follow along with the complicated plots in many of the things I enjoy watching and reading.
Dramatis took way more than six minutes to finish, though the rough version was working in about 10 minutes. I probably spent another 10 hours or so on it, adding features, debugging stuff, and tweaking the design. Dramatis is also not free to use (see below), though the costs are relatively trivial.
Both of these tools use a SQLite3 back-end database, and a web-based front end driven by a small Python web server. They are not suitable for large scale use in this form, but they are perfectly fine for personal use.
These two projects showed me the real power of Claude Code: It's like having a programmer in my house, at my beck and call, and who is capable of turning my "I wish I had a tool that…" thoughts into actual tools, often times in a matter of minutes. And now that I know what it's capable of doing, I have a long list of little things I've wanted to get out of spreadsheets (or Notes or TextEdit, etc.); Claude Code gives me the perfect way to get those projects done.
Downsides
What are the downsides of using Claude Code? First off, I'm obviously not learning how to do any of these things—I'm not getting better at writing code, or designing web pages, etc. But that's OK: I have limited time available for all the things I want to do, and I'm not looking to find a new career as a programmer. I just want a solution that works for me.
Second, I have no idea how good—or bad!—the code is that Claude Code is writing. For simple projects that run on my Mac, don't expose anything to the Internet, and that don't deal with logins or passwords, that's not of great concern to me. If I were planning on trying to sell a solution, or write something with more security implications, it would be of great concern, and I'd probably pay to have someone review and clean up what Claude created.
Another downside (potentially) is cost: Depending on what you do with Claude Code, the resulting project may not be free to use, and could cost some dollars to develop. (Claude Code itself is included in Anthropic's paid plans, and its use counts against your general usage limits).
This is where things get a bit confusing, because a paid Claude account does not give you access to the Claude API, which is how your code accesses Claude's models (if they're required for the task), and which you can use to continue development if you use up your paid account's limits.
It's all about the tokens
To gain access to Anthropic's API, you need both an API key and purchased tokens, which vary in cost to use depending on which model you're accessing. These tokens are then used up as your program queries Claude. But not every project will use Claude—The Queue for instance, is just a local database of TV show information, so it's free to use. (It does retrieve TV series and movie information, but does so using public personal-use API keys that you provide.)
Dramatis, on the other hand, relies heavily on Claude for its data, so it requires tokens. The app isn't scraping web sites to create its character charts, its pulling the data from information it has in Claude's knowledge base. As such, each retrieval costs some number of tokens—a few for the query, and more for the response.
So how much did it cost me to test Dramatis during development, given that I had to go back and forth with Claude Code quite a bit to get the data right? I bought $5.00 worth of tokens, which are used for both input (questions to the model) and output (data from the model).
Overall, I spent 483,832 tokens, with a total cost of $3.47 to develop the system. (This simplifies things a bit, as output tokens cost five times as much as input tokens, but it's close enough.) In other words, not a whole lot of money.
When I make a query to the system to get a new book, movie, or TV series, the number of tokens required varies. Retrieving Season 1 of "24," for instance, required 5.3K tokens, or just a bit shy of $0.04. Retrieving data on all 36 seasons of The Simpsons cost something like $1.50.
But I invoke these costs only once, as the data is then in the local database for future reference. (The tool displays the tokens spent, and their cost, in the title bar after each query.)
Depending on what you build, you may find yourself burning through tokens—and hence, money!—at a fast rate.
Interacting with Claude Code
After some stumbles out the gate, here's what I've found to be the best way to interact with Claude Code (for me, not necessarily for you). I actually start my project in the desktop app, as that interface is designed to be more conversational and general in nature—Claude Code is very programming-focused.
Here's what I asked Claude for when I started the TV Tracker project:
I try to track the TV series I'm currently watching. I'm currently using a couple spreadsheets, as seen in the attached images. I'd like to replace this with a web page where I can see, edit, add, and delete items. I'd like to build it in claude code, but to keep token usage down, data should be stored locally in a sqlite3 database.
The main interface would be a list view, where shows I'm not current on would be highlighted. Adding a new show, it'd be nice if the app could figure out how many seasons are currently out, and whether a new season is planned or not. I should be able to override any entries. I'd also like to be able to just put in a show title, and have the app look up the network, seasons, etc.
Can you write a prompt for this I can give to claude code?
I attached screenshots of the spreadsheet, and Claude then spit out a database schema, and bullet points on features and operation of the system, as seen at right.
Amazingly, it also OCRed the spreadsheets, and used that extracted text to create a comma-separated list of watching and recommended shows for the app to use to populate the database! We went back and forth on a couple of points, but soon, I had a completed prompt.
Making The Queue
I copied everything the Claude app output, then switched to Terminal. I made a new folder for the project cded into that folder, then launched claude. I copied and pasted the input prompt from the Claude desktop app, and six minutes later, it had built the whole thing: The front-end web page driven by the Python server, the logic to manage the records, the web-based data lookup, and the back-end SQLite3 database, pre-loaded with my data:
It worked perfectly on the first try, and I've just refined it and added features over time—movies weren't initially tracked, for instance.
Making Dramatis
On the Dramatis project, it actually started with a TV series from The Queue—Dark, a German series on Netflix that involves time travel. I was having a heck of a time following all the characters, so I asked Claude (the app) for a guide. It presented the data in a very similar manner to what you see in Dramatis, and I found it so interesting and useful, I wondered if it could create a generic version for books, movies, and TV series.
Making Dramatis with claude was much like being a project manager with a programmer on your team. I would notice something in the output—incorrect grouping, duplicate movie titles, not getting all of a book series, etc.—and tell claude what the problem was. It would then go fiddle with the code some more and give me a new version to try. Or I'd ask for some revision to the interface, and it'd come back with rev 2 (or rev 257).
I did a fair bit of work on this one, as I will use it a lot and wanted it to work in a certain way. But still, in under probably ten hours spread over a few days, the final app was ready to go. It does require real-world money to use, but not huge amounts, and the data it provides makes my entertainment more enjoyable.
Wrap up
If you've been looking for a way to create some more-intelligent tools to replace notes or spreadsheets, and wished you knew how to code to do so, maybe you should give Claude Code a try. I've been very impressed with what it's created for me so far. Just be aware it may cost you a bit of money—both during testing and for ongoing use, depending on what you have it write.
(And I promise, full writeups on both are coming, just as soon as I get some time to complete them.)