Recipe 03: AI as juicer
Reviving my kitchen metaphor—this time for app-building AIs and existential asterisks.
If you’ve lived with me, you know I don’t stock juice in my kitchen. Most often I fill my fridge with half-finished condiments; leftover coffee (that I re-market back to myself as “iced”); a thing of milk; homemade stock; veggies from my CSA; and fridge wine a friend left. It means two things are rare in my house. One, it’s hard to make cocktails. Two, I rarely drink the juice.
It’s why it took me until last week to finally sip on AI by vibe coding my way through a problem: I kept missing my friends’ events they posted on Instagram. Below, I walk through a tool AI and I built that solved my problem, what I learned, if and how I’ll continue to use it (hint: like a juicer), and whether what it makes will go sour.
Problem solved in five days
Maybe you too have had this annoyance, but I’m missing many of my friends’ events—deejay gigs, pop ups, readings, you name it. My friends post about them on Instagram, but my feeds’ arteries are clogged by ads, recommendations, bots, and boosted posts. Instagram, originally meant to share photos and updates, is now like flipping through an infinite TV commercial loop.
For years, I craved for an app that would put event information on a single page (or better yet—my Google Calendar), but with zero coding experience, it remained an idea with no tools to build it.
In the past, I had thought of using AI as a sort of proto-Cortana (or Cortana reincarnated?); that is, a virtual chief-of-staff-meets-web-developer-meets-iterator-meets-cheerleader. After futzing with ChatGPT last year, which led to unseasoned copy and LLM hallucinations, I set AI down and worked on other things. But this year’s latest wave of AI updates involve being able to code pretty well and develop working, sharable prototypes. (Two weeks ago, Anthropic announced Claude’s ability to make apps).
Turns out they are much better. Like, solved my five-years-long problem in five days better. I now have a site named Momentum, which you can view here, that daily grabs my friends’ recent Instagram posts and puts them on a single page that I can edit and add to my calendar.
Here’s what I did:
ChatGPT: Guided by my frustration of missing events, I asked GPT to give me a rundown of the event app ecosystem, what competitors do, and how I could build and position this idea as distinct and unique. After an hour or two, we had a serviceable Working Paper (here).
Claude: I gave the working paper to Claude and asked it to build a style guide. Not only did it spit out a pretty good guide (minus the color choices), it also built a navigable in-app prototype using the information it found in the working paper—such as how it should work, what sign ups could look like, and calendar integrations. Claude then suggested I used Vercel (which I’ve already used to host projects like the Post Center Art Center) or Cursor to host it.
Cursor: short-lived (~1 hour test), I tried to give Cursor the Working Paper, Style Guide, and Claude’s prototype code, but after a few hours of me running errands, the AI forgot our chat (and turns out doesn’t autosave). I started from scratch three times until I just gave up and moved on to Vercel’s v0.
v0: Acting like a frontend developer, I sent the same materials I gave to Cursor to Vercel’s v0, and within a few minutes, v0 had a (albeit buggy) demonstrable prototype I could publish and deploy to the internet.
I then flew the URL around and got great feedback from people—the most helpful being to pair the idea down to a Chrome extension that could find my missing friends’ events and put them on a site.
At this point, I had spent just two days on this.
Claude and v0 recommended Make.com and Airtable for the backend—aka, use Make to get daily Instagram information, plop it into an Airtable that v0’s frontend could grab, de-duplicate, and throw up on the website. Then, we grabbed my existing Stripe account to build out a working donation feature.
The remaining three days on this were dedicated to building Make.com and Airtable automation. Turns out, it’s very hard to automatically find event information on sides like Resident Advisor or Gmail, so we stuck to just Instagram, which even that was a challenge. It was a lot of v0, Claude, and I scratching our heads to try and find errors, correct them, deploy them, and repeat.
Eventually, I’d like Momentum to be for others so that they could make a profile and add their friends’ accounts. But for now, I’m basking in a “My god, these apps really did this” energy. One part amazed, one part stunned.
AI as your Juicer, not your Chef
If we go back to our recipe analogy, AI rightfully worries both cultural and white collar workers. It appears weekly - companies lay off junior staff in favor of AI. Yet after playing with these tools myself, I’m not yet convinced every email factory job will go away forever. Even if all of these apps could agentically talk to one another and build something like Momentum in just a day rather than five, it would still need me (a human) to show up with a real world problem, explain why it’s happening, troubleshoot, and shape some potential bets on a solution. (Think Ryan Singer’s Shape Up).
So, no, I don’t fully buy AI will replace us as the chefs in our own kitchen. I’d rather wager they’ll become a tool like emails, computers, and iPhones (Think General Purpose Technology). They appear to rely on humans to experience reality, see and feel and live its challenges, then ask AI to make the world (we hope!) a more durable, equitable place. It begs some questions: will it free up humans to tackle actual challenges instead of being buried in busywork we all despise? Will we let them work for us instead of against us? Or is that just a slippery slope?
For now, I think of AI as a juicer in my kitchen. It takes my raw ideas and juices them into something drinkable. It may not work every time, but until then, it’s been nice to taste the juice it can make and stock it as needed.
The Existential Asterisks
Three arguments come up consistently when I talk about this work to friends and colleagues:
Climate change acceleration: as AI accelerates energy use (Google’s emissions went up by 51% since 2019, and some forecast a data center quintupling), the energy consumed in these companies’ massive data centers won’t be offset if everyone just installed solar panels. It’s a magnitude problem we haven’t seen—ever—and in these days of policy, the efficacy of regulations that force companies to report on data and water usage feel laughable when the next regime change incinerates them.
There’s currently no clear solution to countering this challenge. What is still happening, nevertheless, are market trends that always incentivize companies to work with less for more profit. (Think: AI leaders (“technofeudalists”) are inherently self-interested in growing their capital; that’s the whole point of capitalism.) People who build computers have, since their invention, always worked on reducing their energy use to keep costs low and customers excited. Anything that helps their bottom line, like reducing the number of energy-guzzling data centers, will likely be on the table.
Updated recommendations from carbon scientists release nearly daily. Some AI platforms have already become 25x more energy efficient in just the last year—and 100,000x in the past ten. Maybe right now we’re simply in a bit of a Cambrian explosion in the AI evolution (Category is: the Night of a Thousand AIs) that may level out and even decline to keep costs down.
So, no, I’m not arguing to just “trust” companies to do better. (Peter Thiel, the co-founder of Palantir, recently hesitated on camera when asked if he’d want humanity to endure. Yikes!!!) I’m more so saying the picture is so unclear as to how we’ll solve this that, in the interim, if these AI bots are coming for my (and your) job, we might as well know what they do and what they can’t.
Speaking of coming for your job: while thousands of artists are hard at work fighting for their collective rights against AI, we’re still left with staggering projected job losses and creation numbers from AI disrupting nearly every industry. Some articles say we’ll lose 9 million, gain 11 million; others say we’ll lose 90 million, gain 170 million. It’s all speculation still, sadly.
What I do know is, by using these tools, I got to study their particular weaknesses (v0 needs a lot of manual suggestions for de-bugging; Claude needs to be reminded to be succinct; GPT is still pretty strictly directional (it serves me what I asked, not what I need)). These systems fail; they mess up; they need guidelines, direction, intention, context. And if their energy consumption does in fact plummet, they may be here for a long time, if not become a requirement for most jobs to use.
Linguistic singularity: much like the idea of cultural singularity, we may also worry that the AI-ification of ads, copy, images, etc. could homogenize and flatten language. (Ice cream so good.) And yet, in the face of AI, schools are folding it into curricula while still encouraging criticality, independent strategic thinking, and creativity. Recent research has shown that, even if AIs begin to self-improve recursively (i.e. become increasingly smarter on their own), they seem to still flop compared to a human’s creative abilities. New guardrails also come up every day, like HHH plans that keep AIs helpful, honest, and harmless. I’m cautiously optimistic on this one.
Phew!
I hope you enjoyed. I’ll end this meaty (fruity?) read with something to listen to: the podcast On Being’s unedited conversation with artist, technologist, and philosopher James Bridle. Bridle’s book Ways of Being: Animals, Plants, Machines: The Search for a Planetary Intelligence is on my list; I’ll write about it if I find anything.
Many thanks to several people for their time and feedback on Momentum and this project, including Ritam Mehta, Nikki Vartabedian, Fabiola Belaen, devon parrott, Miguel Senquiz, Brendan Byrne, Erin Keeffe, Doug Sharp, Boaz Sender, and Spencer Duncan.
Whew! That was a good read !! Keep it up -:)