How Olive Young PMs work with AI
더 중요한 고민에 집중하는 올리브영 PM이 AI로 일하는 방식
B · Useful, limited Olive Young tech blog write-up of two product managers' AI routines presented at an internal workshop; concrete steps and time figures, but self-reported by individuals with no team totals or quality checks.
Summary based on the full text·Original: about 12 minutes to read
In short
An Olive Young tech blog post (29 October 2026) relaying cases from the product centre's August workshop. A commerce-display PM automated her daily routine: a scheduled trend briefing, agent-assisted metric checks, weekly metric collection, meeting pages, meeting notes and ticket creation. A coupon and promotions PM, rebuilding the core discount system in a new back-office tool, built clickable prototypes for 7 screens from screenshots and linked the 7 specs so one policy change updates all of them. Both stress that the PM still decides what to change.
Key numbers
- 30 to 10 minutesmorning trend reading after an agent-scheduled daily briefing
- up to 1 hour to 5 minutesweekly Amplitude metric collection per domain, filled into sheets by a scheduled agent and Apps Script
- 30+ to about 10 minuteswriting meeting notes and action items, depending on the meeting
- 7linked specification documents (5 internal, 2 partner screens) kept consistent from one policy table
As reported by the source.
Korean·Web page·opens oliveyoung.tech
What it reports
- Trend briefing: list sources once, prompt for items, send time and channel, check output before wiring it to mail or Slack. Morning reading went from 30 to 10 minutes and no day is skipped.
- Metrics: core metrics and sources connected to the agent; when something looks off, she asks for a week-on-week table with significant moves first, hypotheses, segments to split (channel, device, new or returning) and, if possible, an Amplitude dashboard link.
- Weekly reporting: a skill defines metrics and periods; every Monday a scheduled agent task pulls each domain's figures and Apps Script writes them with formulas and a week-on-week summary; the PM checks and approves. Up to an hour per domain became 5 minutes.
- Recurring meeting pages are cloned automatically with dates updated, cutting about 5 minutes a week to zero.
- Team ticket rules (title format, sections, initial status) are coded as a shared skill, so tickets for each deployment area come out in one format; new colleagues use it for onboarding.
- Prototypes: capture an existing back-office screen for content and a new-system screen for look, then add changes through conversation. Source code was used for the first few, screenshots only after that.
- Spec consistency: numbered markers on each screen capture link to a policy table; changing it in one place updates all 7 documents, including the numbers on the captures.
How they got it
Each PM describes their own setup and time savings. No method for the time figures is given, and the post does not name the agent tools used.
Limits
- Individual, self-reported time savings; no team-wide total.
- No check on accuracy of agent-collected metrics beyond the PM's final review.
- Two people's workflows; adoption across the product centre is not quantified.
Why it matters for companies
The post is a practical catalogue of low-risk PM automations that most product teams could copy within weeks. The more durable ideas are the shared ones: ticket rules turned into a team skill, and specs linked to one policy source so documents stop drifting. Those change team consistency and onboarding, not only one person's minutes.
Key passages in translation
Korean original
도메인마다 최대 한 시간씩 걸리던 집계가 이제는 5분이면 정리됩니다.
Collection that took up to an hour per domain is now done in five minutes.
Korean original
솔직히 말하면 이 '맞추는 일'이 기획 자체보다 더 손이 갔어요.
To be honest, this "keeping in sync" took more effort than the planning itself.
Korean original
만들고 정리하는 건 AI가 하지만, '무엇을 바꿀지' 판단하고 검토하는 건 여전히 PM의 몫이에요.
AI builds and organises, but judging and reviewing what to change is still the PM's job.
Summary based on the full text. Read the original for the full argument and data.Read the original on oliveyoung.tech