Bloom

가장 활발하고 솔직한 AI 커뮤니티

000

When AI starts acting, what should people and companies focus on?

28 May 2026

What was discussed

When AI starts acting, what should people and companies focus on?

Oracle is a database company, yet the keynote was about AI agent memory. Run Claude Code or a multi-agent setup and memory files holding your working context pile up locally, and inside them sit API keys, passwords, even connected internal information. That is fine for personal use, but the moment the tools go to work at a company, critical assets end up scattered across somebody's PC. Oracle's answer was a simple one: keep agent memory in Oracle DB, and security improves while every record is preserved, which makes governance possible.

Lee Kyung-hoon, CAIO of ChannelTalk, spent seven or eight years as a VC before joining, this January, one of the companies he had invested in. He too initially wrote AI off as "a fad that will burn out in two or three years," until a founder in his portfolio told him "AI is about to replace people" and changed his mind. Picking the customer-support market was an investor's call: what LLMs do best is work that has right answers and runs on language, and after coding, that is customer support. Korea's SaaS market tops out at 2 trillion won; the call center market is 10 trillion.

ChannelTalk's AX came in stages. A dedicated AX team failed; what shifted the mood was putting one person who was good with AI on every team. The decisive turn came when they wrote "impact delivered with AI" into performance reviews and handed out company-wide Claude Code accounts. The next day, posts saying "look what I built" exploded, and it eventually led to an unlimited token policy. When they opened the books, non-developers were spending a million won a month and one developer 50 million won a month, but that developer was doing the work of ten people. And the in-house cafe ordering system built with Claude Code? It was built by the barista.

Asked what humans should focus on in the age of AI, he kept his answer short: "We're Homo sapiens. We should focus on thinking." Work that visibly reeks of AI actually scores worse, and something short that keeps only the essentials is worth more than a long, AI-scented document. "People who use AI well," he said, "turn out to be people who are simply good at their jobs." He rejects Dario's end-of-white-collar-work thesis; ChannelTalk chose growth over efficiency, ramping up hiring and committing to a 30% raise in salaries.

The roundtables opened with a question: how far can you trust what AI produces? Across nine tables spanning different industries and levels of experience, the conclusion was strikingly consistent. AI executes; humans make the final call. Only the degree of intervention varied. Finance intervenes heavily because an error does real damage, while LG CNS leaves close to 100% of its coding to AI. The line, in the end, was drawn by how much responsibility a decision carries. Any step that carries no weight of responsibility can be handed to AI.

The subject that drew the most conversation was what to teach the next generation. Kim Sun-woo, a first-year data science student, drew applause for saying that education built around finding predetermined answers is something AI does better, and that what we need is fundamental thinking ability. Several tables agreed that learning how to speak, instruct and delegate, education in natural language, comes before learning to write good Python and JavaScript. And at the end, everyone pointed somewhere similar: the more advanced AI becomes, the heavier problem definition and responsibility become, and in the end what matters most is the relationships between people who meet each other's eyes in the morning.

Read the full write-up

Event video

Gallery

When AI starts acting, what should people and companies focus on?, photo 1

When AI starts acting, what should people and companies focus on?, photo 2

When AI starts acting, what should people and companies focus on?, photo 3

When AI starts acting, what should people and companies focus on?, photo 4

When AI starts acting, what should people and companies focus on?, photo 5

When AI starts acting, what should people and companies focus on?, photo 6

When AI starts acting, what should people and companies focus on?, photo 7

When AI starts acting, what should people and companies focus on?, photo 8

When AI starts acting, what should people and companies focus on?, photo 9

When AI starts acting, what should people and companies focus on?, photo 10

When AI starts acting, what should people and companies focus on?, photo 11

When AI starts acting, what should people and companies focus on?, photo 12

Next event

Proposals that win

Park Ji-hoon of AWS opened with a single idea: TITO, or Trash In Trash Out. Feed a proposal weak information and a weak proposal is all you get back. Only 20-30% of what matters is written in the RFP; the real needs, the other 70-80%, have to be asked for in person. That is why the sales team has to be in the field, and why any need sales fails to collect never makes it into a single line of the proposal. He also warned against the "triangle of death", a document that talks only about your own strengths and skips any analysis of the customer's pain points.

Silicon Valley AI trends you won't find on X

Oh Jae-kyung, EF's country manager for Korea, took the mic first. The obvious question was why a 60-year-old Swedish education company would turn up at an AI community event. He said EF is genuinely wrestling with whether people will still travel abroad to learn a language in an era when AI handles all the translating and interpreting. The point he left the room with was a clear one: language now starts everyone from the same line, and speaking some English is no longer a competitive edge but a tool for meeting someone and opening up a new opportunity.

A sedative for the age of AI FOMO

The first session asked what AI adoption actually looks like inside a conservative organization. Kim Sung-soo of Samsung Electronics' LSI division supports 7,000 people (4,000 of them active users) on 100 GPUs, and built an environment on top of that infrastructure to run Claude Code on their own GPUs. LSI works in RTL, a hardware design language whose code has never been public, so general-purpose models are useless and an in-house tool was the only option. For two or three months he fought executives who wanted the GPUs handed only to "the few who are already good at AI," while he argued that "the people with the domain knowledge are the ones who should be using them." He pushed it through in the end, sending four company-wide emails along the way.

AI Growth Marketing

Guest speaker Choi Je-him called his talk "So What's the Problem?", the question an executive at his first job used to ask him five times over, the five whys. AI is a field where it is hard to claim a right answer, so rather than walk through features that go stale within a week, he said he would share real cases instead. The point is not picking up features but learning how other non-developers solve their own work with AI. The proof: all five award winners at the OpenAI Codex Skillathon he took first place in were non-developers.

AI Summer Night Rave

The first session opened with An Jae-man, CEO of VESSL AI. Behind him was a track record: electrical engineering and mathematics at KAIST, then infrastructure work at Watcha and Cookie Run, where he lived through scaling from 10 million users to 100 million. What pushed him to start a company was, unexpectedly, personal. After his mother was diagnosed with cancer, he said, he turned toward solving medical problems with AI.

Humans and new technology

What prompted the invitation was a YouTube video in which he talked about artists and AI. Rather than pushing AI away, he argued for using it aggressively in production, and for digging deeper into the narrative and context surrounding the work instead. That principle applies not just to music but to every creative field, software included, which is exactly where it connected with the community.