You can check on my intro-to-Claude Code’s capabilities projects which impressed me. This more recent project involves the fact that I just gave birth so items like:
- Is my baby peeing enough?
- Is my baby pooping enough?
- How much is my baby sleeping?
Are top of my mind as is:
- How does the future of agentic coding effect my visibility into the answers to these questions?
Claude Code nearly one shots the solution
Naturally while I was breastfeeding and my baby, happily sated, doted, and gazed upon, fell into a doze, I asked Claude a few questions.
and on confirming yes:
Claude code not working on the phone, I had to wait until I got back to the desktop to try this one out. Prompting Claude Code:
Build a Python service that automatically logs SNOO Smart Bassinet sleep sessions into the Huckleberry baby tracker app in real time.
It went ahead and built it nearly one shot.
Do you even need to know how to code?
I did use my programmer brain thrice:
- I had to copy some credentials into an
.envfile, filling in a.env.example. For this it was helpful to have intuition about where to find dotfiles, as well as the ease of knowing my way around a terminal andcp - I had to be a bit careful that it used a virtual environment to install the python libraries instead of system python when Claude gave me the reigns to get started (bad Claude!)
- and when it was all working (and I tested), I needed to know to ask how to keep it running on my computer, which send Claude on an errand to do so.
I imagine a novice would have figured this out as well through intuition or conversations with Claude. It just saved me some time and made this even more seemless.
Easy Sleep tracking now enabled to feed nap wake window insights
I swear by the Snoo bassinet which makes swaddling, back sleeping, sleep tracking (it autologs sessions), and baby monitoring easy (it alerts you as it notices the baby stirring), and when the baby is older helps connect sleep cycles. Every baby is and family is different; your mileage may vary, but when my 2nd baby was a “I refuse to sleep unless held” baby, this proved a great alternative to keep recommended sleep safety guidelines. It’s original app was simple, just a nice visual of the sleep over a 24 hour period. But they’ve since expanded the app to make that three or four clicks in, with a lot of clutter, and the other logging features leave a lot to be desired.
This is in contrast to the Huckleberry sleep app which also offers an all inclusive sleep/feeding/diaper tracking, but focuses on sleep itself. The tracking interface is nicer – and even though I only log feedings and diapers these days for the first few weeks to ensure everything is on track while otherwise more-sleep-deprived than normal – I much prefer it for those items.
But where Huckleberry really shines is sleep a few months in. It can use the data to help predict when best to set the baby down for a nap, which for certain babies can be extremely accurate (especially as this is sometimes non-intuitive, being far before the baby displays tiredness signs). Sleep tracking manually is a pain though – and since the bassinet does it automatically it is awesome that this project helped me integrate.
With Claude Code, this style of project is so much easier than agentic IDE development just a few months ago
A few months ago I did a cool project using Copilot and VSCode to develop an app that would use the Tesla API to let me know when my husband was on his way home (with his consent, of course!). That involved a few nights of back and forth, and the process of finding the .plist solution to running it locally on Mac was longer. I’m convinced Claude Code could have nearly one shotted this, which is significant improvement.
It was delightful to a few days postpartum be able to brainstorm a connection that didn’t exist, that would make my life easier, and to make it happen in short order. All problems aren’t like this, but this was indeed a “programming shaped problem”. I’ve taught python, around, the, world to beginners and finding “programming shaped problems” is often the hardest part. In fact, when I taught my course in Antarctica I was delighted to have a reliable attendance of 6 (more than 10% of the winterover crew!) all the way up until my final class for final projects which had a dismal attendance of 1. The final project: finding a use for this wonderful tool was intimidating; it was hard for folks without a programming background to brainstorm one independently. Since when teaching, I’ve given a menu of suggestions rather than relying on those new-to-programming
So, as of now I think existing programmers still have an edge in that we see the world already with the “programming shaped problem” lens: “I know just how I’d automate that” or “with all these wonderful tools I know exist, I have ideas about what to build”. I’m looking forward to using that edge to build more things in the coming month. And over time, with the improvement of these tools and the integration of them into all our lives, I see more folks coming with their creativity to stitch and build the world together in new ways with AI
A future of API DJs, tokens as currency, a present of VC subsidized abundance
As a final thought: As MCP servers, and tools like Claude Code are getting more powerful, I see APIs as the first class citizens, with a variety of “API DJs” remixing them into just the UI or connection a user wants. This might be the end user, or it might be a marketplace with some other creative soul having made something to share. The era of a one-sized-fits all application UI and UX may be over, with us all interacting in bespoke ways instead.
Until that point, as people create with AI without a great way to share that is also one-shot (of course we have Github – but it’s not the same), there may be a vast duplication of work and waste of tokens as users reinvent the wheel, or at the very least a lonely bit of software that didn’t find its small set of superfans in the world. So what’s next for Claude Code I’d suggest would be a way to share and click-reuse what’s created (perhaps for a small fee or trade of tokens). In that sense tokens become a currency: burn some to create, earn some when people use your creations, perhaps even sell some to achieve liquidity. That’s in a world where tokens are somehow scarce. But for now – I feel like I’m in the VC subsidized golden era, where I’m being charged nowhere near enough for the value this is able to create. I’m certainly getting my $200 worth this month!
I also see products sharing their backlogs of user requested features that are self contained, and harried product managers outsourcing their implementations to the crowd potentially even putting bounties (in terms of product credits or tokens) on their successfull solutions. Some people see tools like Claude Code causing the death of SaaS services. Rather I think it might be the desert after a long-delayed rain: the backbone of SaaS software enabled by an AI rain to let a thousand features bloom.
Photo credit: Anza Borrego Foundation
You must be logged in to post a comment.