How I Built a SaaS in 3 Months With No Team and No Marketing Budget

I shipped Super Prompts in about three months. No co-founder. No marketing team. No outside funding. Just me, a problem I kept running into, and the stubbornness to build through every wall I hit.

This isn't a "here's how you can do it too" post. This is what actually happened — the decisions that worked, the ones that almost broke things, and the parts I'd do the exact same way again.

The Problem That Wouldn't Go Away

I generate AI content every day. Images for social posts, video prompts for client work, music prompts for projects. And for months, I stored them the way everyone does: some in chat histories, some in Apple Notes, some in random text files on my desktop, a few bookmarked in browser tabs I'd never reopen.

Every time I needed a prompt I'd used before, I'd spend 10 minutes searching through five different apps. Sometimes I'd find it. Sometimes I'd rewrite it from scratch, knowing I had a better version somewhere.

The third time I rewrote a Midjourney lighting setup I knew I'd already perfected, I opened a code editor instead of a notes app.

Starting With the Smallest Thing That Could Work

The first version of Super Prompts was embarrassingly simple. A text field, a save button, a list. No categories. No AI features. No design system. I built it in Next.js because that's what I know, hooked it up to Supabase because I didn't want to manage a database, and deployed it on Vercel because one `git push` and it's live.

The entire MVP took about two weeks. Not because I'm fast — because I was ruthless about scope. Every feature idea got the same question: "Do I need this to save and find a prompt?" If the answer was no, it went on a list I promised myself I'd look at later.

Some of those features are still on that list. That's fine.

The Part Nobody Talks About: Operations

Building the product is the fun part. Keeping it running is where solo founders actually spend their time.

Within the first month, I hit a Supabase egress limit. Five marketing images stored in their public bucket were being served on every landing page visit. AI crawlers were hitting the page hundreds of times, and each visit downloaded 26 MB of assets. I burned through 5.5 GB of bandwidth on a free tier in days.

The fix wasn't glamorous: move the images to Vercel's `/public/` directory where they're served from edge cache for free. Keep Supabase for actual user data. Took an afternoon, but finding the root cause took longer than fixing it.

Then came the bot problem. Spam signups started trickling in through the email form. I added Cloudflare Turnstile — an invisible CAPTCHA that doesn't add friction for real users. But ad blockers kill Turnstile's script, so I built a detection layer: if the security check doesn't load within five seconds, the form shows a message explaining why and offers Google signup as an alternative.

These aren't interesting problems. But they're the problems that determine whether a product stays alive or quietly dies while the founder is off building the next shiny feature.

Building the Marketing Machine Before Having an Audience

I had zero marketing budget and maybe five hours a week for promotion. So I automated everything I could.

Super Prompts now runs a fully autonomous content pipeline. Here's how it works:

1. A draft generator creates social posts across three content types — pain points, hero statements, and tips — each tuned to a different part of the funnel
2. An AI quality check scores each draft against three simulated personas. Anything below 7/10 gets rejected automatically
3. Approved drafts get rendered into branded images using templates
4. A scheduler publishes them through Buffer on a set cadence: pain points daily, hero statements weekly, tips biweekly

The whole thing runs on cron jobs. I don't touch it. The monthly cost for the AI calls that power this pipeline is about $0.34. Not a typo. Thirty-four cents.

I also built a stock-based replenishment system. Instead of generating a fixed batch every week, the system monitors how many unused drafts exist for each content type. When stock drops below a minimum threshold, it generates exactly what's needed. The generator runs daily but only produces content when the bank is low.

Is this overengineered for 25 users? Maybe. But I'd rather spend one weekend building it than every morning thinking about what to post.

The Blog Runs Itself Too

I wanted the blog to support SEO without requiring me to sit down and write a post every week. So I built an auto-generation pipeline.

A scheduled task runs weekly, generates a long-form blog post on a topic relevant to creative AI workflows, assigns it a future publish date, commits it to the repo, and pushes. Vercel auto-deploys. The blog page checks the publish date at render time — if the date hasn't arrived yet, the post returns a 404 and doesn't appear in the index.

The first auto-generated post was about writing AI music prompts. It came out at about 1,200 words with real, usable prompt examples. Not the kind of AI content that reads like it was written by a committee — more like the posts I'd write myself if I had unlimited time.

I still write some posts manually. You're reading one. But having a baseline of content flowing without my involvement means the blog grows even during weeks when I'm deep in product work.

What I'd Do the Same Way Again

Start with a real problem. Not a market opportunity, not a trend, not a "what if." A thing that annoyed me enough to build a solution. That motivation doesn't fade when the code gets hard.

Ship before it's ready. The first version of Super Prompts had no folders, no tags, no search. Just save and scroll. People signed up anyway. The features they actually needed were different from what I would have guessed.

Automate the boring stuff first. Marketing, publishing, monitoring — these are the tasks that steal your best hours if you do them manually. Every system I automated freed up time I spent on the product itself.

Stay on free tiers until you can't. I ran on Supabase's free tier until the egress problem forced an upgrade. That upgrade was a real decision backed by real data, not a premature "enterprise plan just in case." When I did upgrade, I knew exactly why and could calculate the ROI.

What I'd Do Differently

Set up analytics from day one. I went through three analytics setups before landing on one that worked reliably. Vercel Analytics broke, self-hosted Umami failed, and I finally settled on Umami Cloud. If I'd started there, I'd have two extra months of data.

Write the positioning statement before writing the code. I figured out that Super Prompts is "the prompt manager built for creatives" about six weeks in. Everything before that — the landing page copy, the early social posts, the Product Hunt draft — had to be rewritten once the positioning clicked. Starting with a clear sentence about who this is for would have saved a lot of rework.

What's Next

Super Prompts launched on Product Hunt on April 17, 2026 (Vercel Day). The product is live, the content engine is running, and the build keeps going.

If you're a creator who uses AI tools and you're tired of losing your best prompts in chat histories, give it a try. It's free to start, and it might save you the same 10-minute search that made me build this thing in the first place.