Content, AX and SaaS: the new OS of the AI era

What was discussed

Seo Jin-ha, a security manager at Microsoft, opened the evening with a five-minute briefing. 95% of companies already use AI, but only 1% are ready to use it safely. The one thing he kept coming back to was account management: it looks complicated, but locking down a single account stops 99% of security incidents. He walked through a real case in which someone turned MFA off for ten minutes to share a YouTube Premium plan, lost the account, and ended up with 1.2 million won billed to the Gemini API, and he flagged the data-oversharing risk of pasting API tokens straight into a prompt.

Builder Josh's real name is Kim Seung-kwon. He studied design in both high school and university, a designer to the bone, and started out in 2012 on a salary of 24 million won. He taught himself to code, wrote a book along the way, joined SK Telecom, and closed out his corporate career 13 years later earning 150 million won. Standing at the point where the next rung after team lead was the executive track, he asked himself whether that was something he actually wanted to do, and changed course. Through startups and side projects, he found that starting a company was easier than he had assumed.

The conclusion he arrived at was that business is trust, accumulated. It is a line from Chung Ju-yung and also the principle behind Justin Welsh's work. Good content gets you found by the algorithm; longer content lets people go deep, and trust builds up. Rather than making the content he feels like making, he reads the data to work out what the market wants, concentrates on long-form, and runs a funnel that pulls customers from YouTube into Joshua, his AX agency.

Making that content, he ended up reporting on AI-native organisations. GPTers gave an agent called Ppojjagi a name, a voice, and a daily diet of context, raised it like a Pokemon, and handed it the entire operation of a study community tens of thousands strong. Delta Society has everyone, non-developers included, using GitHub and Claude Code as company-wide tools, with SOPs updated daily. SpaceY tells an AI to build something mid-meeting and has a prototype to show by the time the 45 minutes are up. Samyang Foods, instead of trying to change the existing organisation, spun up a separate AI-native team.

Josh's own company, Joshua, runs the same way. Claude Code and Codex are the operating system, an agent called Hermes handles the day-to-day work, and GitHub is the company's single source of truth (SSOT). Meetings get recorded, pile up in Obsidian, and the agents learn the context from there. A Slack agent named Jenna drafts the newsletter and handles tax invoices and B2B email, and at 9 every morning Hermes proposes what the day should look like. "One Day, One Automation" is the operating rule, with a near-term goal of AI taking over 20% of the work, and he was upfront about the limits too. His closing line was unambiguous: "Going AI-native isn't about changing what you have. It's starting over from zero. Begin by recording your meetings."

Across ten roundtable groups, the realities of very different organisations came pouring out, from Samsung, Hyundai, LG, Hanwha, Toss, Lotte Innovate and Fujifilm among others. Lotte Innovate drew a line between greenfield work with no legacy and brownfield work layered on existing systems. Hanwha Vision said that simply confirming they weren't the only ones flailing made the night worthwhile. Fujifilm argued that defining the process, how we actually work, has to come before the technology. Everyone landed in the same place: the technology is already here, and it is process, accountability, literacy, data and culture that haven't caught up. Claude Bloom exists to close that gap.