Moving up the abstraction stack one Slack post at a time (harness details inside)
The “good old days” of hobbyist programming
In the early 2010s I was a iOS developer on the side of being a PhD student. The project I cut my teeth on, my collaborators went on to found Stripe. A passion project achieved national publicity (including an award from then VP Joe Biden). Another project later became Signal, used by millions around the world, and influencer of world events. Suffice it to say, I love a good hobby project, and I have had some great collaborators on home runs on my dance card.
This week I built an explorer through my twitter archive, (prompt shared, as well as the deploy architecture in this post), and I stumbled across some old tweets from my iOS programming days. It was as simple as searching Storyboard on my newly agentically deployed website: https://twitter.christinecorbettmoran.com/. Here I see how the iOS programming ecosystem evolved during that period: the UI portion became gradually more abstract. As someone who has always identified as a backend programmer, this was very welcome.
This was a big step in abstraction, then, but nothing compared to what we have going on now. It seems like things are changing weekly if not daily, so if I’m going to catch the wave of sharing what I do now then well, now’s the time. And, as I went though in The Last Generation of Programmers, I can provide some historical perspective as to why this is so cool. Cue a new blog post.
Spoiler: I do not miss sacrificing my nights and weekends after work to eek out small gains on software I feel must exist in the world. I built that software because I felt I had incredible leverage as a developer, and I felt obligated to use it to improve the world. It was fun, but slow going, even when I was at the top of my game, with people at the top of theirs.
Those were not the good old days: I have a lot more leverage now. These are the great days, and let me show you how I do it this week, and I’ll let you know how it changes over time.
The Harness: OpenClaw + Slack, iterating with Claude Code on my local network
I have a local OpenClaw install (details here) on a Mac Mini which runs my agent, who I call Nova. I don’t personify Nova (in my head I typically just call them “the Claw” or “my AI agent”), but that makes for dry and stilted reading, so it’s Nova hereafter.
For this post I’m going to focus on the case study of the Twitter archive website Nova and I built. That post has a great outline of the prompt as well as the deploy infrastructure choice we landed on, but on reflection, what is probably more useful to the world than sharing the prompt, is outlining the harness behind all that, in bespoke non AI intermediated prose. Because I care. I promise all em dashes were lovingly handcrafted in the em dash region of France by yours truly (there are only a few toward the end, I can’t help it, I have an MFA). In any case, buckle up: here we go.
After some iterations in how Nova and I communicate, I’ve settled in on communicating primarily on Slack. We have one Slack channel for every project, archived when done. I find the suite of Slack channels is a much more intuitive way to keep track of projects than any of the frontier labs’ applications, where things get lost in a sea of chats or their projects are very clunky. Interacting via Slack feels just like a normal work day, for my hobbies. Here’s a sampling of what we have cooking currently:
For the prompting, I do most of my prompting via Wispr Flow from my phone on the Slack channel. When I kick things off, I usually give the agent full run and ask It to serve things locally.
Serving locally means that for this initial step I am able to check out from my phone anywhere in the house and attend to other things while Nova cooks in the background. Then we iterate several times. This is agile and lets me focus on the features I want before worrying about things like deploy infrastructure, security, or performance.
The last step is brainstorming collectively an architecture to deploy. This has a lot of back and forth, and once we settle on something, I typically briefly have to be on console at Nova’s computer to configure some API keys if there are services that we haven’t used before.
What the future holds for my hobby harness
As Nova and I collaborate on more projects, needing to be on console happens less often. I’m optimistic, with things like Stripe projects, that agentic provisioning will get easier and easier, with fewer calls to be on console at my agent. I don’t have access to Stripe projects yet (get in touch Patrick!), but this promises a one-line set up for a variety of services, making getting API keys and an agent set up to use seamless.
Each day the way I interact with Nova evolves. Under my current system, I refine the markdown files, the channels which power it, and the process to maintain context. I’m sure an entirely new system may arise to supplant it, and I’ll be here on the bleeding edge to give it a try when there’s a glimmer of what’s next.
OpenClaw is bleeding edge and fun locally, but each week I see some of its best features incorporated into the frontier labs’ standard enterprise ready offerings. I saw a talk at the [un]prompted conference on the anatomy of an Agentic Personal AI Infrastructure, where the speaker has orchestrated some complex tasks via Claude Code. And now in Claude Desktop you can dispatch tasks from your phone, and with Cowork, allow it to control your computer. This is great news as these great features are shipping, practically tripping over themselves, to reach the masses.
Get to building yourself!
Although things are changing daily, it’s never been a more exciting time to build. Being a hobbyist builder it’s amazing to see spillover in process improvement I get from one project to the next. I saw that when I was working multiple iOS applications, for very different purposes, simultaneously in the 2010s. With higher leverage today, the cross applicability of process also has scaled enormously.
I’m privileged to find postpartum a time of creative flourishing, and—you are never going to believe this—wrote a blog on it. The fact I can kick off an army of agents in one hand while holding my breastfeeding baby with the other is incredible. When I return to work after my entitled parental leave, I expect what I’ve learned from my hobbyist pursuits to make me a better engineering leader and a more effective engineer—especially seeing how fast the bleeding edge features reach the mainstream. But I don’t need potential professional mission success to motivate building. It’s just so much fun to do so right now at this point in history, so if you aren’t already, there is no better time than now.
You must be logged in to post a comment.