The demo took 5 minutes. The actual launch took 3 weeks and $31 a month I didn't fully account for until the second credit card statement arrived.
A while back I wrote about spinning up a working AI prototype in roughly the time it takes to brew coffee. That post somehow became the most-read thing on this blog, which tells me people are hungry for the "you can do this fast" story. Fair enough — you can build fast. But a few readers emailed asking the same thing: okay, but what does it actually cost to go from working demo to something real people can download?
So here's the follow-up nobody asked for but a few people specifically did.
The Prototype Wasn't the Product
The BOGO deals app from episode 10 was the guinea pig here. The AI prototype ran locally on my machine. No server. No database. No store listing. Just a script that made API calls and printed JSON to a terminal. Total cost at that point: whatever electricity my laptop used.
Then I made the mistake of thinking "this could be a real thing." That's where the meter starts running.
The Stack, Piece by Piece
Let me just go line by line because this is the part nobody writes down clearly.
Language model API — $9/month (average)
I'm using Claude Sonnet via the API. The deals app does a classification + short summary pass on each deal that gets submitted. In month one, with basically zero users, I spent $3. In month two, after I posted about it and got a small traffic spike, I hit $14. I'm treating $9 as the rough average, but it's variable and that variability is mildly stressful. I set a hard spending cap after the $14 month.
VPS for the backend — $6/month
Self-hosted on the cheapest tier I could find that wasn't a free plan with artificial throttling. n8n runs here for the automation layer, plus a small Node process for the API endpoints. $6/month is the actual bill, not an estimate. It's been rock solid for about 7 weeks, which I'm slightly surprised by.
Database — $0/month (so far)
Using the free tier of a managed Postgres host. I've got about 340 rows in the main table right now. The free tier limit is 500MB. I'll hit it eventually but not today.
App store fees — $99/year (Apple) + $25 one-time (Google)
This is the one that stings the most on a per-month basis if you're building something small. Apple's $99/year works out to $8.25/month. I keep forgetting to count it as a monthly expense and then it hits the card in January and I feel personally attacked. The $25 Google fee is one-time, so I've already paid it and I'm done.
Domain + SSL — $12/year
Bought a .app domain for the landing page. It auto-renews. SSL is handled by the host for free. Basically $1/month that I genuinely never think about.
Email for transactional stuff — $0/month
I'm using a free tier here too, capped at a few hundred emails a day. The app doesn't send that many. If it ever does I'll have a different kind of problem.
Total: ~$31/month recurring, plus the $25 I already spent once
Building is free. Distribution infrastructure costs money every single month whether anyone opens the app or not.
What I Didn't Expect
The ratio surprised me. The AI part — the thing everyone obsesses over — is actually not the biggest line item. The Apple developer tax runs almost as much per month as the LLM calls, and the VPS costs more per month than the AI does on a slow traffic week.
I kept reading takes that framed AI APIs as the scary cost center. For me, at this scale, it's the app store ecosystem that's the fixed overhead. The AI bill is at least proportional to usage. Apple charges me $99 whether I have 1 user or 10,000.
What I got for $31/month after 7 weeks: 83 installs, 41 monthly active users, and enough usage data to know which feature I built nobody actually uses (the "alert me when this deal expires" notification — 0 opt-ins out of 83 installs, which is its own embarrassing story).
0 out of 83 users opted into the feature I spent the most time building. The receipts do not lie.
The Hidden Costs That Aren't On the Credit Card
Time is the obvious one and I'm not going to lecture about it, but I'll note that the 3 weeks of "actual launch" work broke down roughly like: 1 week of wiring the backend properly, 1 week of app store submission back-and-forth (got rejected once for a privacy policy issue I didn't anticipate), and 1 week of me staring at analytics wondering if 83 installs was good or terrible.
Turns out it's neither. It's just a number.
The rejection loop with the app store cost me about 6 days of calendar time for an issue I could have avoided if I'd read the privacy policy requirements before submission instead of after. That's a receipt too, just not a dollar one.
The app store rejection didn't cost me money. It cost me 6 days, which at my level of impatience felt worse.
Is $31/Month Worth It?
Genuinely don't know yet. The app is free with no monetization because I wanted to learn the distribution side first before layering in payments. So the current ROI is: $0 revenue, $31/month spend, 41 people using something I built.
By the math of a normal business, that's a bad business. But I didn't build the deals app to get rich. I built it to understand what "shipping" actually costs so that the next thing I ship, I go in with open eyes.
The 5-minute demo was real. It genuinely took 5 minutes to get something working with AI at the core. But I'd be doing everyone a disservice if I let that be the whole story. The demo is the easy part. The $31/month is the part you sign up for afterward, every month, whether you feel like it or not.
I've got a note in my calendar to revisit this cost breakdown at month 6. If the user count is still 41, I'll probably shut it down. If it's grown, I'll have a different decision to make. Either way, I'll write it up here.
TL;DR: The AI prototype took 5 minutes; the actual shipped product costs $31/month — and the LLM API isn't even the biggest line item. App store fees and infrastructure add up faster than the AI does.
Next episode: I tried adding a paywall to the deals app at week 8 — here's the conversion rate, and why I almost immediately regretted the timing.
Get the $31 Stack Prompt Pack — $19
The exact prompts, cost calculator, and n8n templates from this stack.
Top comments (0)