How to Build an App With Replit Under Pressure

You can build an app with Replit fast enough to finish before a single Dota 2 match ends. Brazilian developer André Elias did exactly that: roughly 30 minutes of prompting to create a working voting and ranking app, and about 20 more to publish it to TestFlight, all while playing.

What Was the Challenge?

The challenge was to build an app with Replit during a single match of Dota 2, a five-versus-five strategy game where matches average around 40 minutes. André Elias, a Brazilian developer who says he has shipped several profitable SaaS products and logged more than 2,000 hours in Dota 2, set the rules himself: the app had to be functional before the game ended.

He chose a deliberately simple product: a swipe-based voting app for the Brazilian dev-YouTube community, where users like or dislike content creators and a live ranking shows who makes the best content. Simple scope matters here. An agent-built app succeeds or fails on how small and concrete the request is, and a voting list with photos is about the right size for a 30-minute window.

The setup was genuinely hostile to concentration. He recorded the game and his laptop at the same time, Steam took about three minutes to launch, and the matchmaking started while he was still writing his first prompt. He played Juggernaut, the carry hero he normally uses, a role that demands constant farming attention.

How Replit Handles Prompt-to-App Development

Replit is a browser-based development platform with an AI agent that builds software from natural-language descriptions. According to the creator's walkthrough, you describe what you want, and the agent generates the application including database, authentication, and hosting, showing progress in real time in the browser.

The practical flow he described has three parts. First, a prompt describes the app in plain language. Second, the agent builds incrementally and you watch the result live. Third, publishing takes about three clicks once you are satisfied, with no mobile development knowledge required.

That last point is what makes the experiment interesting. Traditional app publishing involves build tooling, signing certificates, and store configuration. On Replit the creator clicked a publish button, generated an icon, and had a live URL before his match finished. The mobile path to TestFlight required only signing in with his Apple account.

What the Agent Actually Built

The first version arrived while the match was already underway, and the creator reports it was genuinely functional: photos of the YouTubers, swipe-to-vote mechanics, and a working ranking page. For a first generation pass with no human code written, that is a real result, and his own reaction on camera was surprise at the quality.

It also shipped with visible defects. Several profile photos were missing or rendered as cartoon avatars because the agent could not find real images, including his own. Some buttons were cut off the screen. And the voting logic had no reset, so a single test vote was permanent during his session, which blocked further testing until he prompted the agent to add a reset button.

The timeline below, reconstructed from the video, shows how the work split across the match:

Where Multitasking Broke Things

The honest lesson of the video is about attention, not tooling. Every time the app needed input, the game punished him. Looking away during the laning phase cost him about five creeps of farm, and early farm deficits compound for the rest of a Dota match. His team lost the first game in 33 minutes.

He worked around the attention problem with the game's own mechanics. Each death gave him a respawn timer of roughly 30 seconds, which he used as focused review windows for the app. It is an accidental timeboxing system, and it is the most transferable trick in the video for anyone juggling an agent-run build with anything else.

Two problems were structural rather than attention-related. Finding real profile photos was the single slowest task in the whole process, slower than building the app itself; he tried LinkedIn images and was blocked by privacy restrictions. And his Replit free trial ran out mid-match, forcing him to enter credit card details while his character stood idle in lane.

Publishing During a Second Match

With the app functional, he started a fresh match and published it live. The web publication completed before that match ended, faster than he expected. The mobile route followed: he signed in with his Apple credentials and the build appeared in App Store Connect and in TestFlight, where he installed it on his phone and voted on real profiles from inside the running game.

His totals for the whole effort: about 30 minutes to create the app, and about 20 more to publish it. Apple's own review process for public App Store listing is a separate step he did not complete on camera; he notes it typically takes around two days. Everything demonstrated in the video, from prompt to installed test build, fits inside two matches of Dota 2, the second of which his team won.

What This Experiment Proves and What It Does Not

The supported conclusion is narrow and useful: for a small, well-specified app, an AI agent platform can take you from idea to a deployed, testable product in under an hour, even with the operator only partially attentive. The scope matters. A voting list with photos is a bounded CRUD application, exactly the shape current agents handle well.

The video does not show production readiness. The app had missing assets, layout bugs, and no evident testing beyond the creator's manual votes. Photo sourcing, an entirely human problem, consumed more time than any code task. And the mid-session credit expiry shows that platform billing is now part of your build pipeline whether you like it or not.

If you attempt something similar, the video suggests three conditions for success: keep the first prompt small and concrete, expect at least one round of visual bug fixing, and schedule your focus windows, because a dying Juggernaut is not a project plan.

FAQ

• Can you really build an app with Replit in one Dota match? In this experiment, yes: the creator reports a functional voting and ranking app after roughly 30 minutes, inside a single match. The app was small in scope, which is the main reason it fit the window.

• How long did publishing take? About 20 additional minutes, by the creator's account. The web version was live before his second match ended, and the mobile build reached TestFlight during the same match.

• Do you need to know mobile development? No. The creator states he published without writing mobile code, using Replit's publish flow and his existing Apple account credentials. Apple's public App Store review, which he did not complete on camera, typically takes around two days.

• What was the hardest part of the build? Finding real profile photos, not writing code. The agent could not reliably locate creators' images, privacy rules blocked LinkedIn photos, and missing or cartoon avatars persisted until the end of the match.

Turn Your Own Videos Into Articles

A 15-minute video like this one holds a complete, messy, honest account of a build under pressure, but only people who press play ever see it. The same thing is probably true of your own content: walkthroughs, interviews, and lessons that live on YouTube and nowhere in writing.

With Skala Blog you can paste a YouTube URL, get a transcription, and generate a structured written article from it, the same way this page was produced. If you have knowledge worth reading, it is worth letting it out of the video player.

Visit Skala Blog, drop in a video link, and turn what you said into something people can find, quote, and skim.

Source video