The firm program an appraiser built with $100-a-month AI (감정평가사가 한 달 100달러 AI로 만든 회사 프로그램)
B · Useful, limited An informal Korean architecture podcast (Digit TV) in which a certified appraiser and two small-practice architects describe first-hand how they build and use AI tools; concrete and credible, but anecdotal with no time-saved or accuracy figures.
Summary based on the full transcript·Original: about 65 minutes to watch
In short
Episode 2 of the podcast "Critic-Eve" on the Digit TV architecture channel (uploaded 8 November 2026, 65 minutes). Lee Sang-hyup, a certified appraiser at Daehan Appraisal with no coding background, built a desk-appraisal tool on public-data APIs that his firm adopted as its official internal site. Two architects from small practices describe using AI for competition studies, layouts and research. The common lesson: start from what is repetitive or slow in your own work and ask the AI how to remove it.
Key numbers
- $100 a monthClaude Max subscription the appraiser uses, enough for one or two websites and three or four apps
- about 15 GBbulk public dataset of apartment official prices he parsed into the internal site
- under 10,000 won a monthhosting cost on the smallest Naver Cloud server, paying only for a public IP
- about three weekstime a competition planning study for a large public building normally takes, now explored with AI layout options
- 27 units / about 50 peoplehanok variants generated for a competition study, and the people shown them
As reported by the source.
Korean·Video·opens youtube.com
What it reports
- The appraiser started on Claude Pro for the Excel add-in in February; after a three-to-four-hour one-to-one session with a founder friend in April he moved to Claude Max at $100 a month.
- Desk appraisal: entering an address now pulls land and building records through public-data APIs, replacing one-by-one lookups. Colleagues liked it and executives made it an official internal site; a recent addition maps redevelopment zones with notices and news.
- Apartment official prices: he parsed the roughly 15 GB public bulk download so one click returns prices for a whole complex.
- Hosting moved from his office PC to the smallest Naver Cloud server, under 10,000 won a month; he asks the AI to fix breakages.
- Architects pay for ChatGPT Plus and buy Midjourney or Kling for single months, about 30,000-60,000 won a month in total.
- Hanok study: 27 variants (form, space, material at 100/50/0%) generated with Claude-written prompts and Midjourney, shown to about 50 people to find where a building stops reading as hanok.
- Competition planning: AI produces two or three editable vector layouts that the three-person team critiques as a jury. More competitions are entered, but thinking time, not drawing, caps the number; output per person is up, projects per month are not.
- Legal and regulatory checks stay with people because an error ends a project.
How they got it
Conversation among practitioners who know each other, summarised from uploader subtitles. Tools are described, not demonstrated in detail.
Limits
- Informal podcast; no time-saved, cost-saved or accuracy figures for the appraisal tool.
- One appraiser's experience; the firm's adoption is stated, not documented.
- Some speaker names and figures come from subtitles and may contain errors.
Why it matters for companies
The appraiser's tool is a common pattern: a domain expert with a $100 subscription turns public data and personal frustrations into an internal system the firm then relies on. That creates value fast and a maintenance and support question later, since the tool lives on one person's knowledge and a cheap server. The architects' experience is a useful caution: AI raised output per person, but the number of projects is bounded by judgement time, and high-stakes checks stay manual.
Key passages in translation
Korean original
그래서 그때부터 100달러를 결제를 해서 맥스 요금제죠. 맥스죠. 고거를 결제해서 지금까지 쭉 쓰고 있습니다.
So from then on I paid $100 for the Max plan, and I have been using it ever since.
Korean original
기본적으로 내가 업무에서 뭐가 불편한가 그쵸. 그거를 가장 고민을 많이 했던 것 같아요.
Basically I thought hardest about what was inconvenient in my own work.
Korean original
법규 같은 거를 만약에 검토를 하더라도 그건 너무 크리티컬한 부분이기 때문에 틀리면 거의 끝장나요.
Even if we review regulations with it, that part is too critical: if it is wrong, it is basically over.
Summary based on the full transcript. Read the original for the full argument and data.Watch the original on youtube.com