What AI can't replace, and what we should grow instead

What was discussed

The opening framed Claude Bloom as "the most active and most honest AI community." Active means gathering good people in a good space and building events worth showing up for. Honest means putting the fear and sense of loss that AI brings, the Blue, on the table alongside the wins, and working through it together. A morning coffee rave and a hackathon in the park were teased as upcoming formats.

MyRealTrip CTO Heo Won-jin opened with customer support. For fifteen years the company handled 75,000 inquiries a month by hiring more people, but early last year it brought in AI. The rollout was deliberate: leave the existing chat solution untouched, layer a chatbot over the simplest inquiries first, then, once results showed up, connect the APIs and let agents take actions on their own. Today the AI's error rate is lower than a human's, and headcount on basic support has dropped from 57 to 28, roughly 50 percent, with the remaining staff moved to higher-tier work handling emotionally charged complaints and urgent cases.

Operations changed too. Translating overseas products, mapping them to the right categories, and organizing them by city was manual work that topped out at 25 listings a week. Now you enter a city, AI analyzes the trends, and products are registered automatically at 4 a.m., up to 5,000 a day. Cities in Africa and the Middle East that no one had ever gotten to are now covered, and Cairo and Monaco have produced real revenue.

JoCoding spoke on solo founding in the age of AI. His 2020 "animal face" test, which told you from a photo whether you looked more like a cat or a dog, hit number one on Naver's real-time search rankings and drew 40 million users in English-speaking markets. Two details stood out: it was a static site on Cloudflare running on-device AI, so server costs were effectively zero, and, in a candid bit of hindsight, he still regrets not adding paid plans at the peak of the virality.

He argued that Sam Altman's "one-person unicorn" is genuinely within reach. His evidence was a benchmark in which a human running a vending machine earned $844 while Gemini earned $4,387, a sign, he said, that AI is starting to outperform people on business judgment as well. Whether you are a large enterprise or a team of one, everyone is wrapping the same models, so what is left is the ability to design. That thinking led to his "product builder factory," which took 280 applicants and accepted 3. Asked how it was going, he owned the failure just as plainly: "Everything put together still comes to less than a thousand dollars." On when to quit your job, his answer was to keep the safety net until you are earning five times your salary.

The roundtable question was what AI cannot replace, and which organizations need to become AI-native. Across 16 tables the stories ranged widely: one team cut its engineering headcount by 60 percent and doubled its output; companies on air-gapped networks said adoption is barely feasible at all, but the insights converged. Responsibility, judgment, intuition, communication. AI says a great many reasonable things and then gets it wrong at the decisive moment, and catching that moment is what people are for right now. There was also broad agreement that nothing changes in an organization until its leaders become AI-native first.