I don't think there has ever been a better time to found a startup than now. If you delegate work to artificial intelligence (AI) agents, you can develop products and grow a business at a pace that was impossible in the past, even with a small team.
Mark Manara, head of startups at OpenAI, said on the 18th at Nextrans 2026 at COEX in Samseong-dong, Seoul, that performance of artificial intelligence (AI) models has improved dramatically over the past nine months to a level incomparable with transfer, making it possible to accomplish in a few days what used to take weeks or months. Manara, who leads global startup business at ChatGPT developer OpenAI, introduced changes underway in the startup ecosystem in a presentation titled Next-generation AI era: from models to agents.
He said today's startup ecosystem is changing at a breathless pace. Manara said the changes over the past six months were an astonishing time of acceleration, distinctly different from the transfer two years, and assessed that the introduction and spread of AI agents that carry out tasks on their own are changing how startups operate.
In particular, Manara explained that the main driver of change is that the pace of AI advancement has accelerated exponentially over the past nine months, making it possible to implement high-performance AI agents. A year ago, it was difficult and complex to build AI agents that plan on their own and use tools to perform tasks, but now startups can easily implement AI agents that will take over development work by leveraging state-of-the-art AI models, he said, adding that as a result, the startup ecosystem is witnessing agents surge at a Cambrian explosion level.
To that end, OpenAI has focused over the past year on improving models' reasoning ability, long-context handling, coding, tool use, instruction following, computer use, and document and presentation generation. Accordingly, agents built with OpenAI's latest model, GPT-5.5, have reached a stage where they can perform long-running tasks without human intervention, he said. Last August, agents built with GPT-5 performed tasks on their own for an average of seven hours without developer intervention, but with GPT-5.5 that has increased to 24 hours, he said.
Manara emphasized that as the AI industry moves from the chatbot era, represented by ChatGPT, to the agent era, the growth formula for startups has also changed. He cited the case of AI coding startup Cursor, which SpaceX, the space company led by Tesla CEO Elon Musk, recently acquired for $60 billion (about 91 trillion won). Early this year, Cursor successfully built a web browser by leveraging hundreds of AI agents based on GPT-5.2.
To build a properly functioning web browser used to take months, but Cursor built one in a week using only agents without human intervention, Manara said. During that period, the hundreds of agents that Cursor ran in parallel wrote a total of 3 million lines of code, generated more than 1,000 files, and even developed its own rendering engine from scratch in Rust, a relatively challenging programming language, he said.
As startups enter a stage where they delegate a substantial part of development work to AI agents, even small teams can dramatically boost productivity. Developers now don't assign just one task to an agent but run 10 to 20 tasks in parallel, Manara said, calling it the reason startups can develop products and services and scale their businesses at a speed incomparable with the past, even with small headcounts.
Regarding Korea, he said it is one of the countries that adopt AI technology the fastest and use it most actively, and at the same time it has one of the most dynamic startup ecosystems in the world. He added that he highly values the potential and growth prospects of Korean startups and will continue to invest in Korea's startup ecosystem through the startup support team established in Seoul.
Both OpenAI and rival Anthropic have set up offices in Seoul over the past year and are targeting the domestic startup market. Startups adopt and use AI quickly, making them ideal testbeds for AI companies. Startups are among OpenAI's most important customers, Manara said, because they are the first to try the latest models and tools from OpenAI and then tell us what needs improvement, playing a key role in model research and post-training.