From individual AI to company AX: Buzzni
개인의 AI에서 회사의 AX로 | AI 비용 49% 줄이고 매출 50% 성장한 버즈니 이야기
B · Useful, limited Buzzni CEO Nam Sang-hyup gives first-hand revenue, staffing and cost figures for a company-wide AI reorganisation, with the operating rules behind them; self-reported, and the second half demonstrates the company's own product.
Summary based on the full transcript·Original: about 24 minutes to watch
In short
A talk at KES 2026 (published 16 October 2026) by Nam Sang-hyup (남상협), CEO and co-founder of Buzzni, a Korean search and recommendation company that also runs a home-shopping aggregator app. After revenue fell by close to 6 billion won, the company reorganised around AI: about half of the staff ran the existing service while the rest started new AI businesses. Revenue went from about 6 billion to 9 billion won with fewer staff. The second half demonstrates SayCode, Buzzni's own agent-management platform.
Key numbers
- 60억 to 90억 wonrevenue in Korean won, from about 6 billion to 9 billion (+50%), with fewer staff
- 49%reduction in AI spend claimed in the title (about 50% in the talk) through visibility, routing and per-person model allow-lists
- 1-2people who should take most work from start to deploy to measurement, to cut hand-offs
- $200monthly Claude plan paid for every employee at first, which led to low and then uneven spend
As reported by the source.
Korean·Video·opens youtube.com
What it reports
- Four organisational rules: cut hand-offs so that most work goes from start to deploy within one person, two at most; remove role boundaries (no front-end/back-end split, PMs build and deploy); back the strongest AI users first; and the CEO must drive it personally.
- An "AX team" left alone does not work; even with the CEO pushing, it only barely works. The slowest person in a chain becomes the bottleneck.
- Results: revenue up from about 6 billion to 9 billion won (+50%), staff numbers down, losses sharply reduced.
- Company-level problems with individual AI use: paying every employee a $200 Claude plan led to low use at first, then rising and uneven spend; prompts and results stayed on personal machines and left with departing staff; use across vendors could not be audited in one place.
- SayCode features shown: token and cost views by person and team, routing each prompt to a model of suitable difficulty, per-person model allow-lists, app building by non-developers with MCP connections to company data, account revocation on exit and audit logs. He claims about 50% cost reduction from these measures.
- Rollout advice: audit current use, set rules, pilot in one department, then go company-wide with leadership driving it.
How they got it
The CEO's own account. No method, period or baseline is given for the revenue or cost figures.
Limits
- Self-reported; the revenue rise may have causes other than the reorganisation.
- The cost saving is attributed to the company's own product, which the talk sells.
- No headcount figures are given, only that staff decreased.
Why it matters for companies
The organisational rules carry more weight than the tool: fewer hand-offs, merged roles and visible top-level sponsorship. The cost section is a practical checklist for any firm that has handed out AI subscriptions: spend visibility per person, model routing, allow-lists and keeping prompts as company assets rather than on laptops. The revenue figure is a claim to note, not to benchmark against.
Key passages in translation
Korean original
저희가 60억 매출에서 이제 90억으로 매출이 50% 이상 성장을 했고요. 또 중요한 부분은 인력은 오히려 ... 좀 줄었습니다.
Our revenue grew from 6 billion to 9 billion won, more than 50%. And the important part is that headcount actually ... went down somewhat.
Korean original
여러 사람이 일하다 보면은 그 가장 느린 사람이 이제 병목이 되기 때문에
When several people work on something, the slowest person becomes the bottleneck.
Korean original
그 권한을 가지고 있는이 대표가 아주 강하게 드라이브를 걸고 ... 해야 이게 약간 잘 될까 말까입니다
Only when the CEO, who holds that authority, drives it very hard does it maybe, just barely, work.
Summary based on the full transcript. Read the original for the full argument and data.Watch the original on youtube.com