Kim Do-gyun, head of the homegrown artificial intelligence (AI) startup Dalpa, on the 27th of last month at the company's headquarters in Gwanak District, Seoul, predicted the future of work that AI will reshape. The diagnosis is that the trend of replacing not only simple repetitive tasks but also core white-collar areas such as marketing and planning with AI is entering full swing. This concern directly prompted Dalpa to scrap its previous business structure and switch to an agent-centered corporation.
Born in 1999, Kim founded Dalpa in 2023 with classmates from Seoul Science High School and Seoul National University. The team had conducted AI projects together since their school days, and in the early startup phase developed general-purpose AI solutions, but they changed course after feeling firsthand on actual corporate sites the limitation that AI does not make money.
Global research institutions are pointing in the same direction. McKinsey, in a 2023 report, analyzed that 60% to 70% of working hours could be automated with Generative AI, and Goldman Sachs likewise projected in the same year that about 300 million jobs could be affected by AI. As a considerable portion of white-collar work becomes subject to automation, the assessment is that a trend has begun in which corporations' workforce structures and decision-making methods themselves are changing.
In response, Dalpa chose a strategy of digging deeply into specific industries rather than general-purpose AI. In the early days, it was a project-based solution corporation that carried out AI projects for various corporations, but about six months ago it completely overhauled its business model. It now focuses on building an agent-based operating system (OS) in which AI performs companywide work for consumer-goods corporations such as K-beauty, fashion, and food and beverage. Dalpa has raised 13.3 billion won to date and secured more than 200 client companies.
An agent refers to a system that goes beyond simple Generative AI and, when given a goal, designs and executes the required tasks on its own. For example, when given the goal to increase sales, it automatically carries out a series of processes from selecting influencers and running ads to performance analysis and product planning.
This structure is translating into real results. Dalpa said that in an experiment where the same budget was allocated to a human marketing team and an AI agent, the agent posted performance 3 to 5 times higher. The gap appeared not only in views but also in contribution to sales.
Dalpa organizes corporate data like a single map, then builds a structure in which multiple AIs divide roles and work simultaneously on top of it. An AI that runs ads and an AI that plans products move organically to produce outcomes. COBRA is a system that executes this process by breaking it into small task units and assembling them, creating the optimal combination depending on the situation. Thanks to this structure, the company said it can continuously improve performance without people intervening at every step.
AI inevitably delivers results faster than human organizations because it repeats experimentation and optimization around the clock based on data, Kim said. He added that corporations that adopt it keep increasing their usage, and because the results are visible, the response is not to cut back but to use more.
Dalpa presents as its core vision a structure in which AI actually makes money for corporations. Kim said that if conventional AI was a tool, agents are closer to an organization, and that going forward AI will take on most execution work while people will focus on setting direction and granting approval.
As agentic AI models that perform tasks autonomously, like Anthropic's Claude, advance rapidly, an environment is emerging in which corporations can adopt AI immediately without separate development. Accordingly, the focus of competition is shifting from which model is used to the ability to apply it to real work and tie it to results.
At this juncture, Dalpa's strategy is to preempt the agent-based operating system (OS) by leveraging data and execution structures tailored to consumer-goods corporations. Kim said that a structure will emerge in which one person does the work of 10, and ultimately an era will come in which one person uses agents to generate 10 billion won in sales. He added that corporate organizations will be reorganized from people-centered to structures that combine AI and people. The following is a Q&A with Kim.
— Dalpa: what kind of company is it? How has it changed from the previous model?
In the beginning, it was a company that carried out AI projects for various corporations. The focus was on functions such as chatbots and marketing automation. But this approach stopped at improving some tasks and had limits in changing entire corporations. We have now shifted to a company that builds an agent OS for consumer-goods corporations. The core is a structure in which AI performs corporations' marketing, product planning, and overall operations.
— Why focus on the consumer-goods industry?
Consumer goods is the area where AI effects translate to sales the fastest. Decision cycles are short, and data-driven optimization connects directly to results. In the end, we saw that AI must be a money-making technology, and by that standard consumer goods was the most suitable industry.
— How do results compare with a human marketing team?
On the same budget, there was a 3 to 5 times difference. The gap appeared not only in views but also in actual contribution to sales. Agents are structurally different because they repeat experimentation and optimization around the clock based on data.
— What is the response from client companies?
Corporations that adopt it keep increasing their usage. Because the results are immediately visible, the response is not to cut back but to expand further. Satisfaction is especially high at the management level.
— Is the change that people only confirm realistic?
It is already changing that way internally. Agents produce results and people review them. Going forward, execution will shift to AI and people to the role of approval.
— What are your goals going forward?
Our first goal is to become the best agent OS company in the consumer-goods space. After that, we plan to make a structure in which one person earns 10 billion won a reality and expand into global markets.