Honestly, when I saw the headline “Someone in Korea used AI to cut the prep time for a dose of Korean herbal medicine from 300 minutes to 5,” the first thing that caught my eye wasn’t “whoa, robots can make herbal medicine now.” It was how they did it—because they happened to get right the one thing most people get wrong when they think about applying AI.

What Onerve Did

Let’s start with the facts. There’s a Korean startup called Onerve (오너브), backed by the Korea Institute of Oriental Medicine, working on automating the manufacturing of Korean herbal medicine (한약).

Their system is called HAP. It connects AI with electronic medical records (EMR) to automate the entire flow—from prescription input, to manufacturing, cleaning, packaging, and inventory management. The key is the raw material: they use standardized, freeze-dried herbs in a “cartridge” format—turning herbs that used to require on-site boiling and heavy manual labor into uniform, standardized modules.

The result: prep time for a single dose of Korean herbal medicine dropped from around 300 minutes to around 5. They won a CES Innovation Award and closed a Series A round of roughly 6.2 billion won.

And they’re not alone—another Korean company, Camelotech (with its Cameleon system), is doing almost the same thing and also showed up at CES. So “Korean herbal medicine automation” is turning from a one-off experiment into an actual category.

What I’m Actually Paying Attention To Isn’t the Speed—It’s Which Layer They Automated

If all you take away from this is “300 minutes became 5,” you’re missing the most important part.

When people see AI moving into an industry with a thousand-plus years of tradition behind it, the gut reaction is usually panic: “Are even Korean medicine doctors about to get replaced by AI?” But if you look closely at what Onerve actually automated—it’s the manufacturing, not the diagnosis and prescribing.

Deciding which medicine a person should take, how to adjust the dosage, how to read their constitution—the parts that require judgment and hard-won experience—that’s still the doctor’s job. What the machine took over is the part that comes after the prescription is decided: the repeatable, standardizable step of producing the medicine to spec.

That division of labor is the genuinely smart part of this whole thing.

This Is Basically What I Do Every Day Running AI Agents

I live in Korea, and every day I’m running a whole team of AI agents. The longer I do this, the more convinced I am: whether AI adoption actually works has nothing to do with “can it replace a human” and everything to do with whether you’ve clearly separated “the repeatable layer” from “the layer that needs judgment.”

And here’s the part most people miss—which is actually the real bottleneck in all of this. People assume Onerve’s hard part was “getting a robot arm to measure out herbs.” But think about it—the actually hard part was taking messy, inconsistent raw herbs that come out a little different every time, and turning them into uniform, standardized “freeze-dried cartridges.” The real work of automation was never the machine. It’s standardizing whatever you’re handing off into an interface with a clearly defined spec.

Only once the cartridge-format raw material was standardized could the machine take over.

I run into the exact same bottleneck running agents: the hard part is never “telling the agent to go do it.” It’s that I first have to define a repetitive task’s inputs, outputs, and spec clearly enough—precisely enough—that the agent can actually pick it up. Defining that interface is the real work. So what makes Onerve smart isn’t the machine—it’s that they were willing to first sit down and standardize the “medicine” into a cartridge, and only then hand manufacturing to the machine, while leaving diagnosis and prescribing—the judgment calls—to the doctor.

More Valuable Than Speed: “Every Dose Is the Same”

Following that thread further, the surface-level value Onerve delivers is speed, but what’s actually valuable is consistency.

With traditional hand-made medicine, the concentration and cooking process can vary batch to batch due to human factors. Standardized cartridge-format raw material makes the potency of every single dose predictable and reproducible. In a medical setting where mistakes aren’t acceptable, predictable consistency beats the occasional brilliant outlier by a mile. The exact same principle applies to how you should think about using AI: what matters about an AI doing work for you was never “how impressive was its best output”—it’s “does it reliably deliver at a certain bar, every single time.” Only once it’s reliable can you actually trust it with real work.

What This Means for You

I know for a lot of people, “AI helping Koreans make herbal medicine” sounds like it has nothing to do with them. But the real signal in this story is directly relevant to your day-to-day work: AI adoption lands fastest and most solidly wherever the work is “repeatable and standardizable.”

So if you want to start using AI to actually get things done for you, don’t rush to find “one all-powerful AI that can replace my entire job.” Instead, lay out your work and ask yourself two questions:

  1. Which layer is repeatable and standardizable? (Things you do every day, with fixed steps and a clear spec.) → This is the layer you should hand to AI first. But remember—it’s worth the effort to write the spec out clearly before you do.
  2. Which layer requires judgment and experience, where getting it wrong is costly? → Keep this layer for yourself for now, and just let AI assist.

Get clear on these two layers, and you’ve got the first real trick to using AI well. If even a thousand-year-old medical tradition can be split this way—half handed to AI, while the half that most needs a human stays firmly with a human—your own work will have that same line somewhere too.

Sources