South Korea's overseas direct-purchase market surpassed ₩6 trillion in 2024. Within that market, one startup has emerged that processes ₩10 billion in cumulative transaction volume with just four operations staff. This is the story of Modpick, a cross-border commerce startup. Typically, a commerce company generating ₩10 billion in revenue needs a team of dozens — customer service, logistics, merchandising, operations managers, and more. Modpick is betting that formula no longer holds. Because it built a system where AI agents directly handle product sourcing, customs clearance, and customer service, the math on headcount itself has changed.
The reason this case immediately prompts soul-searching for solo entrepreneurs is simple: the question of "how much of my own work can I hand off to an agent" is no longer hypothetical.
Where AI Moved In First
Modpick's chosen arena is cross-border personal shopping — a service that handles the entire process on behalf of Korean consumers buying from overseas retailers, from local purchasing to customs clearance, shipping, and customer service. This market has three core bottlenecks: product sourcing (figuring out which products from which countries will sell in Korea), customs paperwork (the complexity of country-specific tariff classification codes and regulations), and customer service (language barriers and the repetitive work of handling returns and refunds).
All three have traditionally been considered domains that require highly skilled staff: an eye for good products, knowledge of customs regulations that vary by country, and the ability to handle customer service in foreign languages. To secure these three capabilities, commerce companies build out teams. Modpick went the opposite direction — designing a system where AI agents handle these three areas, with humans making judgment calls on top of that layer.
In product sourcing, data-driven demand forecasting narrows the candidate pool. In customs, tariff-code identification and document generation are automated. In customer service, multilingual response and return-handling workflows have been standardized. As of 2026, Modpick has raised seed funding from Kakao Ventures and Ewha Womans University's technology holding company, and has announced plans to launch an autonomously operated commerce platform in the second half of this year.
What matters more than the funding news itself is what the combination of "four operations staff, ₩10 billion in cumulative sales" implies. If these figures hold up, they signal that a substantial share of the tasks in commerce operations once assumed to require a human touch could be reclassified.
What You Miss If You Take the Number at Face Value
It's an impressive figure, but a few things deserve scrutiny.
First is how the figure is defined. Cumulative sales and annual revenue are not the same thing. If the ₩10 billion is a cumulative total since founding, annual revenue could be much smaller depending on the time span involved. Funding coverage tends to present numbers in their most flattering form, and it hasn't been disclosed how many years it took to accumulate that ₩10 billion.
Second is what "AI handles it" actually means in practice. Even if agents generate customs documents and respond to customer inquiries, that doesn't mean no human reviews the final output. Customs errors lead to fines and held shipments; customer service errors translate directly into platform penalties and return disputes. If what the four operations staff actually do is review agent output and handle exceptions, the structure isn't so much "humans aren't needed" as "humans work at higher density."
Third is the nature of the industry itself. Unlike manufacturing or services, personal shopping doesn't require holding inventory directly — items are sourced only once demand materializes. It's also an industry with a relatively high share of patterns amenable to repetitive processing. Customs codes and shipping conditions vary by country, but the rules themselves are documented. That's a structurally favorable environment for agents to learn from.
Even so, none of these caveats undercut the core of Modpick's case. It remains true that highly repetitive, rule-based work can be handed off to agents, and that this experiment is already underway. The question has shifted from "is it possible" to "which parts of my own work qualify."
What Solo Operators Should Do First
Abstract Modpick's operating structure and one pattern emerges: break work down into information gathering, rule application, and output generation, then hand those stages to an agent. Product sourcing is a process of collecting data and recognizing patterns. Customs processing is a process of codifying regulations and drafting documents. Customer service is a process of classifying requests and generating standardized responses.
All three tasks are repetitive and can be substantially codified into rules. And that characteristic isn't unique to commerce.
For a content director, that might include keyword research, competitor analysis, and part of the drafting process. For a solo PM, it could mean meeting-note summaries, draft feature specs, and status reports. For a one-person founder, it might be inquiry triage, contract drafting, and the first stage of lead generation. The question is simple: what share of my day consists of the flow of "gather information, apply rules, deliver output"?
This sorting exercise should come before picking an agent tool, not after. You first need to map out which stages of your workflow are highly repetitive and which require contextual judgment. One reason Modpick appears to have applied agents effectively is that it drew this line clearly. Agents handle the customs paperwork itself, but the judgment call of "does it make business sense to bring this product into the Korean market" still belongs to a person.
Rule-based processing without context goes to the agent; contextual judgment without fixed rules stays with the person. Drawing that line at the level of individual tasks is the first job solo entrepreneurs need to do right now. I'd like to call this process "task dissection" — taking your own work apart before you adopt any tool.
Modpick's four people review what the agents produce, decide which new categories to expand into, and negotiate with investors and partners. What you do with the time freed up after handing off repetitive work is what determines each operator's value. The person who catches an agent's mistake by reading context, who picks up on market signals the data doesn't show, who extracts the real problem from a customer's complaint — these roles are hard to automate with rules. And they will likely stay that way for some time to come.
If you don't design that role in advance, adopting the tools just adds confusion instead of productivity. The wider the territory AI covers, the more the quality of judgment outside that territory defines the human role. Before adopting an agent, figure out first what kind of judgment you'll be exercising in the space left behind once the repetition disappears.
