The drawing-tools keyword cluster gets tens of thousands of monthly searches with keyword difficulty scores near zero. That single data point drove a micro SaaS in 24 hours: a sketch-to-refined-drawing app built with Next.js and an AI code editor, deployed on Vercel announced on LinkedIn the next morning.
What Does It Take to Build a Micro SaaS in 24 Hours?
A micro SaaS in 24 hours takes three things: a demand-validated idea found through keyword research, an AI-assisted build on a proven stack, and a fast deployment path to a live domain. Brazilian developer André Elias ran exactly this experiment in a video published on July 9, 2025, and reached a working launch in about three hours of actual work, well inside the 24-hour limit.
He set one boundary at the start: the goal was launch, not the first paying customer. He treated the release as a demand test, promising to share sales figures on his social channels if any came in over the following weeks. The video was sponsored by Abacus.AI, whose Code LLM code editor he used for the build, and he framed the project partly as a stress test of how far AI assistance can carry product development.
The whole run followed three phases, which is also the structure worth copying:
### The Three-Phase Timeline at a Glance
| Phase | Time spent | Output |
|---|---|---|
| Idea and keyword research | ~30 minutes | Low-difficulty "drawing" keyword cluster |
| Build with AI code editor | ~90 minutes | Functional sketch-to-drawing generator |
| Deploy and launch | Next morning | Live domain on Vercel, LinkedIn announcement |
Total hands-on time was roughly three hours. Everything else in the 24-hour window was sleep and the overnight gap before the launch post.
How Do You Find a Micro SaaS Idea With Real Search Demand?
You find a micro SaaS idea with real demand by filtering keyword research tools for high-volume terms with near-zero keyword difficulty, then building exactly what those searchers want. This was the first phase of the 24-hour build, and it took about half an hour using Ahrefs, which he noted is a paid tool with free alternatives available.
His method in the video: open Keywords Explorer, set a country (Brazil and the United States both worked), then push the keyword difficulty (KD) filter down from 20 to 5 and finally to 0. For contrast, he showed that ranking for "ChatGPT" sits at 97 out of 100 difficulty with no realistic chance for a new site, while a term like "WhatsApp Web" draws roughly 75 million monthly searches in Brazil, according to the Ahrefs figures shown on screen.
Browsing the lowest-difficulty results surfaced the "drawing" cluster: terms like "drawing ideas", "flower drawing", "cat drawing", and "dog drawing", all with meaningful volume and very low competition. He then used an include filter for the word "drawing" to map the whole cluster before committing to a product.
The product decision followed directly from searcher intent. People searching drawing terms want reference material, so the tool would let a user sketch a rough shape, add an optional prompt, and generate a more polished illustration. Every generated image would become SEO content, aggregated onto pages that target the same low-difficulty queries that inspired the product.
How Was the Product Built With AI in 90 Minutes?
The product was built in about 90 minutes using the Code LLM editor from Abacus.AI, a VS Code-based environment with integrated chat and agent modes, running a Next.js application on localhost. This is the phase where a micro SaaS in 24 hours either happens or stalls, and the AI tooling carried most of the implementation work.
For the model he chose Gemini 2.5 Pro, citing its reputation for programming tasks, and worked primarily in agent mode, giving broad prompts and letting the tool generate the code. Image generation went through the OpenAI API, which required creating a key in the OpenAI dashboard and wiring it into environment variables.
The first functional version existed 31 minutes into the build. A rough sketch of a dog next to a tree produced a misread first generation, but after adjusting the prompt the output matched what he had in mind. He then switched to chat mode to have the AI refine his own prompt engineering, because the raw generations were too polished for the product's purpose: users need images they can attempt to redraw themselves.
One fix absorbed real time on day two: a credit-purchase bug where balances were not updating. He verified the fix by buying credits, watching the total go from 6 to 13, generating an image, and confirming the balance dropped to 12. Small verification loops like this are what turn an AI demo into a launchable MVP.
Where Did the AI Tooling Break Down?
The AI tooling broke down whenever prompts got too open-ended. On day two, while preparing for deployment, the agent started inserting random console logs and code he would never have written, a failure he attributed to giving mega-broad prompts instead of narrow, scoped tasks.
The editor itself warned him about this: long conversations reduce the model's precision, and smaller scope produces better results. His own workflow outside the video is the opposite of what the challenge demanded. He normally writes code himself and calls in AI for specific, bounded chunks, using it to save typing rather than to architect the product.
His conclusion was blunt and worth quoting: learn to program if you want to ship a product alone, and use AI as a complement rather than the center of everything. A non-developer hitting the same hallucinated-code loop would have had no way to diagnose or repair it. He also noted spending credits on corrective agent runs, so vague prompting costs money as well as time.
How Do You Deploy and Launch a Micro SaaS the Same Week?
You deploy a micro SaaS by running a clean production build, pushing to GitHub, importing the repo into Vercel, and attaching a domain, then announcing where your audience already is. That was the entire launch path in this experiment, completed the morning after the build.
The sequence he followed: verify the Next.js build succeeded locally, create a public GitHub repository called draw-with-ai (public deliberately, since he has no long-term plans for the project and invited viewers to continue it), import it as a new Vercel project, set production environment variables, and buy the domain directly through Vercel. He acknowledged Vercel's domain prices run higher than buying elsewhere and pointed out the alternative: configure DNS manually at another registrar. He paid for convenience given the deadline.
Before posting, he caught a missing step: responsiveness. A round of mobile fixes went out, a short demo video was recorded, and the launch post went to LinkedIn, the one channel where he already had reach. He listed Reddit, TabNews, and TikTok as other launch surfaces and framed LinkedIn as a bug-hunting test with real users rather than a marketing push. Design quality was an explicit non-blocker: he judged the interface acceptable because the core functionality, sketch in and refined drawing out, already worked.
What Are the Real Lessons From a 24-Hour Micro SaaS Build?
The real lessons from this build are about validation order, scope discipline, and developer judgment. Keyword-first ideation removed the biggest startup risk, demand uncertainty, before a single line of code existed. The Ahrefs data shown in the video, high volume paired with keyword difficulty near zero, gave the product an audience before it had a name.
Time estimates deserve honesty. The headline says 24 hours; the work took about three, but that figure depends on a developer who could fix build errors, repair a credits bug, and rein in a hallucinating agent on demand. For creators exploring this format, Gustavo Dev Doido's ecosystem, including the Crazystack Typescript course at crazystack.com.br and the Bootcamp do Dev Doido, teaches the underlying SaaS-from-scratch skills this challenge assumed.
The most transferable lesson is the scoping rule: AI coding tools excel at narrow tasks and degrade badly on open-ended ones. Anyone attempting a micro SaaS in 24 hours should plan for short, verifiable agent runs, a manual review of everything generated, and a launch defined as a demand test rather than a finish line.
FAQ
- Can a non-programmer build a micro SaaS in 24 hours with AI? The creator behind this experiment argues no. AI code editors drift into broken code on open-ended tasks, and fixing that requires reading and correcting real code. He recommends learning programming fundamentals and using AI as a complement.
- How long did the build actually take? About three hours of work: roughly 30 minutes of keyword research, 90 minutes of AI-assisted coding to a functional version, plus deployment, a domain purchase, responsiveness fixes, and the launch post the following morning.
- What stack did the project use? Next.js for the application, OpenAI's API for image generation, the Code LLM editor from Abacus.AI with Gemini 2.5 Pro as the coding model, GitHub for version control, and Vercel for hosting with a domain bought through the platform.
- How was the SaaS idea validated before building? Through Ahrefs keyword research: filtering for search terms with high monthly volume and keyword difficulty near zero, which surfaced the "drawing" cluster, then designing the product to serve and capture that existing search demand.
- Was the product finished at launch? No. It was an MVP with a known feature backlog, including a dropdown to choose image models. The launch existed to test whether real demand existed behind the search volumes, not to ship a complete product.
Turn Your Own Build Videos Into Articles
This article followed the same shape as the video it came from: a validated idea, a build, a launch, and the lessons in between. If you have recorded your own process, tutorials, or project walkthroughs, that same structure is already sitting inside your uploads as watchable but unsearchable knowledge.
Skala Blog turns a YouTube video into a written article: paste the URL, the video gets transcribed, and you get a structured draft you can review and publish. For anyone documenting builds like this one, it puts the written version of your work within reach without writing it from scratch.
Fork this article
Start a new branch from the same video, shaped your way. You keep the credit; the original keeps the attribution.
A fork in another language is filed as a translation of this article, so the two pages point at each other. You can unlink it later from the editor.
0/240
You are creating
- Format
- For
- Language
- Source
- Your angle
You will be asked to sign in before it is generated.
Buy credits