[The Era of Solo AI Startups] From Idea to App: Non-Experts Build Apps with Only AI

Now AI Handles Coding Too
'Lean Startup' Acceleration Expected
"A Startup Earthquake... Opportunities for Everyone"

"I never imagined that someone like me, who couldn't write a single line of code, would end up creating an artificial intelligence (AI) service alone."

Lee Sunmyung, Product Manager responsible for product planning at Kakao Mobility, struck gold by developing 'Mirror,' an AI web service that evaluates photos. Mirror is a service that analyzes photos uploaded by users and scores them based on composition, lighting, color, focus, and other criteria. After launching in March, the service gained popularity through word of mouth on social networking services (SNS), with more than 78,000 photos uploaded within a month. That amounts to 2,800 photos per day, or over 100 photos per hour, with users requesting AI analysis. Lee's hobby was limited to taking and editing photos; he had never received formal coding education. He stated, "I am now directly creating the service I had only imagined using AI and improving its features based on user feedback."

The first screen of the website for 'Mirror,' an AI-based photo evaluation service developed by Lee Sunmyung, Product Manager at Kakao Mobility. Official Mirror homepage

Integration with OpenAI Services... The Emergence of AI Photo Analysts

It is now becoming a reality to easily create AI services without a background in computer science. This is expected to accelerate the so-called 'lean startup' approach, which reduces initial startup costs. 'Lean' means 'without excess, slim' in English. The idea is to minimize unnecessary elements or waste based on an idea, maximize efficiency, and operate a business quickly and effectively. The approach involves rapidly launching a minimum viable product (MVP), observing market reactions, and then immediately revising or discarding the product if necessary. Lee Hanjun, CEO of the brand marketing startup Stackers, said, "The ability of entrepreneurs to leverage AI will become even more important," adding, "A seismic shift in the startup ecosystem is about to begin. What is clear is that there are opportunities for all of us."

The rise of lean startups has been made possible by code development support platforms. The reason Lee Sunmyung, who does not have a developer background, was able to create the 'Mirror' service was largely thanks to the platform 'Replit.' Replit is a coding platform that is easy to use for everyone from beginners to experts. Lee subscribed to a paid plan costing $20 per month and created the desired service in a conversational prompt format. Through conversations with Replit AI, he built the basic framework of a webpage where users could upload photos for AI evaluation. He then connected the 'OpenAI Vision API,' which specializes in analyzing image and visual data, and instructed it, "You are a photo expert. Score each category and write an overall review." After completing and releasing this initial version, the service quickly spread by word of mouth on SNS, and the number of users wanting their photos evaluated increased significantly. This fulfilled his long-held dream of creating a service for photographers. He further increased user engagement by creating AI photo critic personas with different characteristics and allowing users to share the evaluation results externally.

Recently, most AI startups have been emerging in the field of visual intelligence, like Mirror. According to a comprehensive survey conducted by the Software Policy & Research Institute in the second half of last year of over 2,500 (2,517) domestic AI companies with at least one employee, the most common AI technology field was visual intelligence, accounting for 33%. This was followed by intelligent agents (17.4%) and language intelligence (13.6%).

Maintenance Is Essential... Concerns Over Unclear Responsibility

Launching a service easily and at low cost with AI is not the end of the story. As the user base grows, continuous maintenance and management become necessary. Lee said, "I devoted myself to 'Mirror' as if I were a full-time founder and developer," adding, "After work, once my baby was asleep, I spent every night analyzing logs of program operations, errors, and user behavior, and refining the design and user experience."

Since paid AI models are used, costs are also incurred. In the case of Mirror, every time a user requested an AI analysis of their photo, Lee had to pay OpenAI, and as the number of requests increased, he was hit with a 'bill shock' of about 400,000 won (approximately $290) per month. He said, "Thousands of photos were being uploaded daily, and payment notifications were going off every five minutes, resulting in unexpected costs," adding, "If the number of users increases further, I could add advertisements to the service, but since it is not yet generating revenue, the burden was significant." As a result, Lee switched to Google's 'Gemini 1.5 Flash' AI model, which offers better cost-effectiveness than OpenAI.

Industry experts agree that building sophisticated and complex services still requires the help of professional developers, and that preparations must be made for unexpected errors or defects. Kim Deokjung, adjunct professor at Sookmyung Women's University, said, "When companies introduce AI coding, their development capabilities may weaken, or responsibility may become unclear."