Written with AI assistance from the source video. The video is the authority.
Log in to leave a comment
No posts yet
When a CEO drops an order in the morning meeting to automate company-wide tasks with AI, a planner or marketer with less than 3 years of experience feels their mind go blank. It's overwhelming enough figuring out what tools to pay for right away, and articles are making a fuss about jobs disappearing.
A McKinsey report announced that generative AI could replace 60-70% of knowledge workers' tasks. However, actual worker survey results published by the Federal Reserve Bank of St. Louis in 2025 tell a completely different story. The time spent using AI on the job was only 1-5%, and the time saved per week averaged a mere 2.2 hours. This is why trusting flashy demo videos and hastily throwing AI into practical work often results in working overtime just to inspect the lies made up by the AI.
Generative AI is not a tool that finishes the work for you. It is closer to an auxiliary device that quickly creates a draft at an 80% level. The remaining 20% of context coordination and fact-checking still belongs to the practitioner. According to OpenAI's 2025 corporate research, practitioners who used the tool daily reduced their average daily time by 40 to 60 minutes, and 75% directly tackled unfamiliar tasks like coding or data analysis. If you expect 100% automation, you will be disappointed, but if you use it for drafting, work speeds up.
| Task Category | AI Domain (80% Draft) | Human Domain (20% Completion) |
|---|---|---|
| Task Details | Data summarization, initial copy writing, basic SQL query generation | Numerical cross-checking, brand tone and manner review, reflection of internal policies |
| Time Proportion | 20-30% of total work | 70-80% of total work |
| Output Nature | Draft subject to verification | Deployable work asset |
To bring management's excessive demands into reality, you must first define the limits of the technology in numbers.
The first thing to look at when choosing a tool is not features, but data security policies. Personal paid accounts like ChatGPT Plus or Claude Pro may use input data for model retraining. Putting internal planning documents or customer response data as-is can lead to a leakage accident. On the other hand, OpenAI and Anthropic's Team plans fundamentally exclude data from model training and provide SOC 2 Type 2 certification and data-at-rest encryption.
| Check Item | Personal Plan | Team Subscription (Team) | Developer Console API |
|---|---|---|---|
| Data Training Status | Included by default (manual opt-out required) | Excluded by default (No Training) | Excluded by default (No Training) |
| Data Retention Period | Permanent storage | Configurable from 30 to 90 days | 30 days or immediate deletion |
| Account Management | Individual payment | Administrator console, member control | API key-based control |
According to Tech.co's 2026 small business survey, companies that spent an annual budget of $1,001 to $2,500 (approx. 110,000 to 280,000 KRW per month) were most likely to save 6 to 10 hours per week. On the other hand, places that only used various low-cost tools costing $10 a month had savings of less than 2 hours per week.
Once the proposal passes, proceed sequentially with Week 1 security setup, Week 2 prompt organization, Week 3 parallel work, and Week 4 time-saving data measurement.
Prompts that worked well on test screens often break when entering practical work. This is because practical data is messy and has many exceptions. Hallucinations making up non-existent numbers, misinterpretation of table data line breaks, and missing instructions for long text inputs are representative examples.
| Stage | Check Content | Passing Criteria | Countermeasures if Failed |
|---|---|---|---|
| Stage 1: Security | Inclusion of customer personal info, unreleased sales data | 0 cases of sensitive info, confirmation of training exclusion | Mask names and numbers before input |
| Stage 2: Error Rate | Results of random test on 30 samples | Factual error rate of 10% or less | Add examples (Few-shot) to prompt |
| Stage 3: Inspection Cost | Time taken to modify output | Within 30% of original writing time | Reduce task scope to draft level |
| Stage 4: Contingency Plan | Alternatives when API goes down or abnormal output occurs | Feasibility of immediate manual switch | Prepare backup manual |
Tasks that pass the checklist are operated in a 5-stage flow.
Listing things like trying out a few tools does not help your career much. What remains is the experience of defining problems, managing risks, and reducing time.
If you reduce simple repetitive task time, the remaining time must be spent elsewhere. Reduce simple writing time from 12 hours to 4 hours, and allocate the secured 8 hours to GA4 data analysis, customer interviews, and landing page conversion rate experiments.
In months 1-3, reduce the time of 1 of your tasks, and in months 4-6, share proven prompts and security rules within the team. In months 7-12, expand the scope to work collaborating with other departments. This is the sequence of moving from a practitioner knocking out simple repetitive tasks to a planner designing work flows.
