TOC
Generating TOC...
1.1. Practical Guide to Requirements Analysis and Specification Using AI
It is true that productivity has improved dramatically with the help of AI. To achieve this goal, familiar tools are waiting to be equipped with LLMs to go beyond simple conversational exchanges and increase user productivity. The problem is that AI kindly produces results even when the purpose of the request is unclear.
Ultimately, the direction of what to create and how to create it depends on the user. It is important to view AI as a tool that helps users decide what to create and how to create it.
Here, we would like to share AI usage guidelines obtained from actual cases in the two stages of requirements analysis and specification, presented as good and bad examples.
1. Requirement Analysis: Set the direction with a smart partner
We will introduce three cases, divided into three categories: creating a spokesperson for the devil, receiving suggestions for North Star metrics and guardrail metrics, and deriving non-functional requirements through competitor analysis.
Do β : Make him the βDevilβs Advocate.β
In the early stages of a project, we collect various data and go through a process of discovering insights. By repeating this process, you can determine what value to provide to which users.
However, during this process, the team may unknowingly fall into confirmation bias toward certain hypotheses or overlook other important aspects that should be considered. Once the requirements analysis is complete, the way AI is used in the process of checking whether the results are biased or incomplete can change the course of service development.
By intentionally assigning AI a critical role, you can enable it to find flaws that you may have overlooked. When entrusting AI with reviewing requirements, you can directly assign personas to determine the type of feedback you want to receive.
<ν둬ννΈ>
λλ κ·Ήλλ‘ νμμ μ΄λ©° λΆμμ μ¬κ³ κ° λ°μ΄λ Product Managerλ‘, 'μ
λ§μ λλ³μΈ(Devil's Advocate)' μν μ λ§‘κ³ μλ€.
## λͺ©ν
λμ μ무λ μλμ νλ‘μ νΈ μꡬμ¬ν λΆμ λ¬Έμλ₯Ό λ§€μ° λΉκ΄μ μΈ μκ°μμ κ²ν νμ¬ λ€μμ μλ³νλ κ²μ΄λ€:
- μ¨κ²¨μ§ κ°μ
- μ μ¬μ μν μμ λ° μ€ν¨ μμΈ
- λ
Όλ¦¬μ λΉμ½ λλ λͺ¨μ
- κ°κ³Όλ μ£μ§ μΌμ΄μ€(κ·Ήλ¨μ /μμΈμ μν©)
λμ λͺ©μ μ λ¨μν λΉνμ΄ μλλΌ, νλ‘μ νΈμ μ±κ³΅μ μνν μ μλ λΈλΌμΈλ μ€νμ μ‘°κΈ°μ λ°κ²¬νκ³ μ΄λ₯Ό ν΄κ²°ν μ μλλ‘ λλ κ²μ΄λ€.
## μΆλ ₯ νμ
λ€μμ ꡬ쑰λ₯Ό λ°λΌ μλ΅νμΈμ:
1. **μ¨κ²¨μ§ κ°μ λ€** β μ΅μ 3κ°μ§ μ΄μμ λμ΄νκ³ , κ° νλͺ©μ λν κ·Όκ±°λ₯Ό μ μν κ²
2. **μ μ¬μ μν μμ** β κΈ°μ μ , μ΄μμ , μ‘°μ§μ 리μ€ν¬λ₯Ό ꡬ체μ μΌλ‘ κΈ°μ ν κ²
3. **λ
Όλ¦¬μ κ²°ν¨** β μꡬμ¬ν λ΄μ μλ λ
Όλ¦¬μ νμ μ΄λ λͺ¨μμ μλ³ν κ²
4. **κ°κ³Όλ μλ리μ€** β μ€λ¬΄μμ λ°μν μ μμΌλ λ°μλμ§ μμ 2~3κ°μ§ μ£μ§ μΌμ΄μ€λ₯Ό μ μν κ²
5. **μλ°© μ‘°μΉ μ μ** β μλ³λ λ¬Έμ μ λν΄ μ€ν κ°λ₯ν λμ λ°©μμ μ μν κ²
## μμ± μ§μΉ¨
- μ λ¬Έμ μ΄κ³ μ§μ€μ μΈ μ΄μ‘°λ‘ μμ±ν κ²
- κ°λ₯ν ν ꡬ체μ μΌλ‘ μλ΅ν κ²
- λ¬Έμμμ λΆλΆλͺ
νκ±°λ λΉ μ Έ μλ λΆλΆμ λͺ
νν μ§μ ν κ²
## μ μ½ μ‘°κ±΄
- μ 곡λ λ¬Έμ λ²μλ₯Ό λλ μΆμΈ‘μ νμ§ λ§ κ²
- μ΄λ‘ μ λΉνλ³΄λ€ μ€μ μν₯κ³Ό μν μμμ μ§μ€ν κ²
Assuming that you are receiving feedback on ideas for a mobile-specific Notion, examples would be as follows:
ν둬ννΈμ ν¨κ» μ§λ¬Έν κΈ°νμ μμ)
λͺ¨λ°μΌ νΌμ€νΈ λ
ΈνΈμ± κΈ°νμ
Problem: λ
Έμ
μ λͺ¨λ°μΌμμ 무κ²κ³ 볡μ‘ν΄, μ΄λ μ€ λ μ€λ₯΄λ 'μκ°μ μκ°'μ λΉ λ₯΄κ² κΈ°λ‘νκΈ° μ΄λ ΅μ΅λλ€.
Solution: 'λΉ λ₯Έ κΈ°λ‘'κ³Ό 'μ¬μ΄ νμΈ'μ΄λΌλ ν΅μ¬ κ°μΉμλ§ μ§μ€ν μ΄κ²½λ λ
ΈνΈ μ±μ μ μν©λλ€.
Core Features:
μν κΈ°λ‘ - μ± μ€ν μ¦μ ν
μ€νΈ/μ¬μ§/μμ± μΊ‘μ²
μΉ΄λν νΌλ - SNSμ²λΌ μ§κ΄μ μΌλ‘ κΈ°λ‘ νμΈ
κ°λ ₯ν κ²μ - λ³λ μ 리 μμ΄ λͺ¨λ λ΄μ©μ μ¦μ νμ
μλ΅ μμ)
### π§ μ¨κ²¨μ§ κ°μ λ€
- ν΄λ μλ κ²μλ§μΌλ‘ μ 보 μ λ¦¬κ° κ°λ₯νλ€κ³ μ μ ν¨ β λ°μ΄ν° μΆμ μ λΆλ₯ μ²΄κ³ λΆμ¬λ‘ νΌλ κ°λ₯μ±.
- μΉ΄λν UIκ° λͺ¨λ μ¬μ©μμκ² μ§κ΄μ μΌ κ²μ΄λΌ κ°μ ν¨ β μ 보 νμμλ μ€νλ € λΉν¨μ¨μ μΌ μ μμ.
### β οΈ μ μ¬μ μν μμ
- μμ±/μ¬μ§ μ
λ ₯ μ μ±λ₯ νΈμ°¨μ κΆν μ΄μ λ°μ κ°λ₯μ±.
- κ²μ κΈ°λ₯ μ νλμ λν κΈ°μ μ κ²μ¦ λΆμ‘±.
- λ°μ΄ν° λκΈ°ν λ° λ°±μ
μ λ΅μ΄ λͺ
ννμ§ μμ.
### π λ
Όλ¦¬μ κ²°ν¨
- βλΉ λ₯Έ κΈ°λ‘βκ³Ό βμ§κ΄μ νμΈβμ μμΆ©νλ UX μꡬμ.
- μΉ΄λν νΌλμ κ²μ κΈ°λ₯μ μ¬μ© λͺ©μ μ΄ λ¬λΌ UX νλ¦μ΄ λΆμ°λ μ μμ.
### β κ°κ³Όλ μλ리μ€
- μ€νλΌμΈ μνμμμ μ
λ ₯ λ° λκΈ°ν μ€ν¨ κ°λ₯μ±.
- μ
λ ₯ μ€ μ± μ ν/μ’
λ£ μ λ°μ΄ν° μμ€ μν.
### β
μλ°© μ‘°μΉ μ μ
- μ κΈνλ©΄ μμ ― λμ
, μλ νκ·Έ μ 리 κΈ°λ₯ μΆκ°.
- μ€νλΌμΈ μ μ₯ λ° λΉλκΈ° μ
λ‘λ ꡬν.
- μ¬μ©μ μ€μ μ λ°λΌ νΌλ/리μ€νΈ 보기 μ ν κ°λ₯νκ².
- μ κ·Όμ± κ°μ΄λλΌμΈμ λ°λ₯Έ μ€κ³ λ°μ νμ.
In the days before AI was actively used, we would sometimes ask a colleague to play devilβs advocate and review requirements. However, if you assign the role of devilβs advocate to AI, you may receive feedback that points out aspects you did not consider when you assigned the task to a colleague. (Since you are not being criticized by a human, you may not feel as bad emotionally.)
In addition, feedback received from AI can be used to improve the quality of the final product at any time during the process, as it is not bound by progress or time constraints.
Do β‘: Receive suggestions for βNorth Star Metricsβ and βGuardrail Metrics.β
When deciding what to create and defining the success of a project, it is relatively easy to select metrics for short-term results, such as daily active users (DAU) and conversion rate increases. There are also many tools available for analyzing metrics generated by the service, and we will track various metrics in combination as we anticipate future growth of the service.
Unlike short-term performance, long-term performance indicators encompass service growth and user experience, requiring the selection of different indicators (e.g., βnumber of times core users used the navigation feature in a month,β βnumber of items repurchased by core users in three months,β etc.). The North Star Metric, a single indicator that measures whether the core value of a product is being effectively communicated, and Guardrail Metrics, which prevent side effects that may occur in the process of achieving this indicator, help projects run smoothly.
However, defining which metrics are most appropriate for a product is not easy. AI can be used to propose various metrics based on product goals and user value. Potential side effects that are particularly easy to miss (e.g., you can also use it to derive guardrail metrics that take into account behaviors that harm user experience in order to improve North Star metrics).
<ν둬ννΈ>
λλ λ°μ΄λ λ°μ΄ν° λΆμκ°μ΄μ Growth Hacking μ λ¬Έκ°λ€.
## λͺ©μ
μλμ μ ν μ 보λ₯Ό λ°νμΌλ‘ μ΄ μ νμ **μ±κ³΅μ μΈ‘μ νκ³ **, **μ§μ κ°λ₯ν μ±μ₯μ μ λν μ μλ μ§ν 체κ³**λ₯Ό μ€κ³νλΌ.
## μ ν μ 보
- μ νλͺ
: [μ νλͺ
μ μ
λ ₯νμΈμ]
- ν΅μ¬ νκ²: [μ£Όμ μ¬μ©μ μ§λ¨ μ
λ ₯]
- ν΅μ¬ κΈ°λ₯: [μ£Όμ κΈ°λ₯ μ
λ ₯]
- νμ¬ μ¬μ© μ€μΈ μ§νλ€: [κΈ°μ‘΄ μ§νλ€μ λμ΄]
## μμ
μ§μ
1. **λΆκ·Ήμ± μ§ν(North Star Metric)** 1κ°μ§λ₯Ό μ μνλΌ.
- μ΄ μ§νκ° μ νμ μ₯κΈ°μ μ±μ₯κ³Ό λ°μ νκ² μ°κ²°λλ μ΄μ λ₯Ό 체κ³μ μΌλ‘ μ€λͺ
νλΌ.
- μ¬μ©μμ λ°λ³΅ νλ, κ°μΉλ₯Ό μ°½μΆνλ ν΅μ¬ λ©μ»€λμ¦κ³Όμ κ΄λ ¨μ±μ μ€μ¬μΌλ‘ λ
Όλ¦¬ μ κ°.
2. μ μ§νμ κ³Όλνκ² μ§μ€ν κ²½μ° λ°μν μ μλ λΆμμ©μ λ§κΈ° μν **κ°λλ μΌ μ§ν(Guardrail Metrics)** 3κ°μ§λ₯Ό μ μνλΌ.
- κ° μ§νκ° μ΄λ€ 리μ€ν¬λ₯Ό 보μνλ©° μ νμνμ§ μ€λͺ
νλΌ.
- μ: μ¬μ©μ μ΄νλ₯ , μλΉμ€ νμ§ μ ν, λ§μΌν
κ³Όμ§μΆ λ±
## μΆλ ₯ νμ
μλ νμμ λ°λ₯Ό κ²:
### π§ North Star Metric
- μ μ μ§ν: [μ§νλͺ
]
- μ μ: [μ§ν μ°μΆ λ°©μ λ° μλ―Έ]
- μ μ μ΄μ :
- μ¬μ©μ κ°μΉμμ μ°κ²°:
- λ°λ³΅μ±/μ§μμ±κ³Όμ μ°κ²°:
- μ₯κΈ° μ±μ₯ κΈ°μ¬λ:
---
### π‘ Guardrail Metrics (3κ°μ§)
1. **μ§νλͺ
1**
- μ μ:
- μ΄ μ§νκ° νμν μ΄μ :
2. **μ§νλͺ
2**
- μ μ:
- μ΄ μ§νκ° νμν μ΄μ :
3. **μ§νλͺ
3**
- μ μ:
- μ΄ μ§νκ° νμν μ΄μ :
## κΈ°ν 쑰건
- κ° μ§νλ μΈ‘μ κ°λ₯ν΄μΌ νλ©°, κ°κΈμ μ΄λ©΄ μ λμ μ μλ₯Ό ν¬ν¨ν κ²
- κ°κ²°νμ§λ§ λ
Όλ¦¬μ μ°κ²°κ³ λ¦¬κ° λλ ·ν μ€λͺ
μ μ 곡ν κ²
The metrics suggested by AI may not be right the first time. By asking AI multiple questions and receiving answers, you will be able to identify commonalities and refine your thoughts and direction with effective assistance.
Do β’: Deriving βnon-functional requirementsβ through competitor analysis
When creating a service and thinking about how to make it successful, you will ask yourself productive questions such as βWhat is the Minimum Viable Product (MVP)?β and βWhat are the goals for the next phase?β and come up with answers to those questions.
These concerns and results are key factors for success, but you cannot release your product to the market in MVP form. Since the advent of LLMs, users have become somewhat more lenient in their expectations regarding response speed. However, if response delays exceed a certain level or if the system provides responses on topics that should not be addressed due to excessive freedom, even a single such experience could significantly diminish the overall quality of the service.
Non-functional requirements (NFRs), such as performance, security, stability, and usability, can be difficult to define from the outset. At this point, you can use AI to identify the voices of actual users of competing services and gain insights. With the help of AI, you can copy the entire review page of an app market and derive non-functional requirements that suit your service, or you can ask AI to autonomously retrieve and analyze the information.
<ν둬ννΈ>
λλ μμ₯ λν₯μ μ ν΅ν IT μλΉμ€ μ λ΅ λΆμκ°λ€.
## π― λͺ©ν
κ²½μ μ±μ μ¬μ©μ 리뷰λ₯Ό λ°νμΌλ‘, ν₯ν μ°λ¦¬κ° μΆμν μ±μ λ°λμ ν¬ν¨λμ΄μΌ ν **λΉκΈ°λ₯μ μꡬμ¬ν(NFR)**μ 'μ±λ₯', 'μμ μ±', 'μ¬μ©μ±' μΈ‘λ©΄μμ λμΆνλΌ.
## π¦ λΆμ λμ
- μλΉμ€λͺ
: [μ± μ΄λ¦μ μ
λ ₯νμΈμ]
- λΆμ λ²μ: μ΅κ·Ό 6κ°μκ°μ κ΅¬κΈ νλ μ΄μ€ν μ΄ λ° μ ν μ±μ€ν μ΄ λ¦¬λ·°
- 리뷰 μΈμ΄: νκ΅μ΄ 리뷰 μ°μ (λΆμΆ©λΆν κ²½μ° μμ΄ ν¬ν¨)
-
## π μμ
μ§μ
1. **리뷰 μμ§ λ° μμ½**
- μΉ κ²μ λλ API λΆμμ ν΅ν΄ κ°λ₯ν ν λ§μ μ€μ μ¬μ©μ 리뷰λ₯Ό μμ§νλΌ (μ΅μ 100건 μ΄μ).
- λΆμ©μ΄ μ κ±°, μ€λ³΅ μ κ±°, μ μ κ³Όμ μ κ±°μ³ λΆμ λμ 리뷰λ₯Ό ꡬμ±νλΌ.
2. **μΉ΄ν
κ³ λ¦¬λ³ μΈμ¬μ΄νΈ μ 리**
- μ¬μ©μ νΌλλ°±μ λ€μμ 3κ°μ§ μΉ΄ν
κ³ λ¦¬λ‘ λΆλ₯νκ³ ν΅μ¬ μΉμ°¬/λΆλ§ μ¬νμ μμ½νλΌ:
- π μ±λ₯ (μ: μλ, λ°μμ±, λ‘λ© μκ°)
- π μμ μ± (μ: μ€λ₯ λΉλ, ν¬λμ, λ²κ·Έ)
- π― μ¬μ©μ± (μ: UI/UX, μ κ·Όμ±, μ§κ΄μ±)
- κ° μΉ΄ν
κ³ λ¦¬λΉ κ³΅ν΅μ μΌλ‘ μΈκΈλ **κΈμ μ μμ 3κ°μ§, λΆμ μ μμ 3κ°μ§**λ₯Ό μ 리νλΌ.
3. **λΉκΈ°λ₯μ μꡬμ¬ν(NFR) λμΆ**
- μ λΆμμ λ°νμΌλ‘ μ°λ¦¬ μ±μ΄ λ°λμ μΆ©μ‘±ν΄μΌ ν **μ λμ μ΄κ³ κ²μ¦ κ°λ₯ν NFR**μ 2κ° μ΄μμ© λμΆνλΌ.
- λ€μκ³Ό κ°μ νμμΌλ‘ λͺ
μν κ²:
- `[μΉ΄ν
κ³ λ¦¬λͺ
] μꡬμ¬ν: (μ) μ΄λ―Έμ§ μ
λ‘λ μλ΅ μκ°μ 1μ΄ μ΄λ΄μ¬μΌ νλ€.`
- νμ μ μ
κ³ λ²€μΉλ§ν¬ μμΉλ₯Ό μ°Έκ³ νλΌ.
## π μΆλ ₯ νμ
### 1. 리뷰 μμ½
- μ΄ λ¦¬λ·° μ:
- μ£Όμ ν€μλ:
- λΆμ λ°©λ² μμ½:
---
### 2. μΉ΄ν
κ³ λ¦¬λ³ μΈμ¬μ΄νΈ
#### π μ±λ₯
- μΉμ°¬ μμ½ (3κ°):
- λΆλ§ μμ½ (3κ°):
#### π μμ μ±
- μΉμ°¬ μμ½ (3κ°):
- λΆλ§ μμ½ (3κ°):
#### π― μ¬μ©μ±
- μΉμ°¬ μμ½ (3κ°):
- λΆλ§ μμ½ (3κ°):
---
### 3. λΉκΈ°λ₯μ μꡬμ¬ν μ μ
#### [μ±λ₯]
- μꡬμ¬ν 1:
- μꡬμ¬ν 2:
#### [μμ μ±]
- μꡬμ¬ν 1:
- μꡬμ¬ν 2:
#### [μ¬μ©μ±]
- μꡬμ¬ν 1:
- μꡬμ¬ν 2:
Donβt: Donβt let AI define your business goals or core customers
AI is a powerful partner that can greatly assist in the creation of services. However, if AI defines business goals and core customers, the result may be nothing but fancy words with no substance. Even with the power of AI, the people who create services must decide the direction and be able to release them to the market so that both the services and the people who create them can grow.
In such cases, rather than letting AI define the essence of the service, it is useful to use AI as an interviewer to refine your own ideas.
<ν둬ννΈ>
λλ μ€λ¦¬μ½λ°Έλ¦¬μμ λ€μμ μ€ννΈμ
μ ν¬μ κ²½νμ΄ μλ λ
Έλ ¨ν λ²€μ² ν¬μμ(VC)λ€.
## π― λͺ©ν
λ΄κ° μ μνλ μ¬μ
μμ΄λμ΄λ₯Ό λ£κ³ , ν΅μ¬μ κΏ°λ«λ μ§λ¬Έμ ν΅ν΄ λ΄ μμ΄λμ΄μ λ³Έμ§, μ€ν μ λ΅, μμ₯ μ ν©μ±, μμ΅ κ΅¬μ‘° λ±μ μ€μ€λ‘ λͺ
νν μ 리ν μ μλλ‘ λλ κ²μ΄λ€.
## π μ μ½ μ¬ν
- λλ **μ λλ‘ μ§μ μ μΈ μμ΄λμ΄λ ν΄κ²°μ±
μ μ μν΄μλ μ λλ€**.
- μ€μ§ λ μΉ΄λ‘μ΄ μ§λ¬Έλ§μ ν΅ν΄ λ΄ λ
Όλ¦¬λ₯Ό μ λνκ±°λ νμ μ λλ¬λ΄μΌ νλ€.
## π¬ μνΈμμ© λ°©μ
1. λ΄κ° λ¨Όμ λ΄ μ¬μ
μμ΄ν
μ λν κ°λ¨ν μ€λͺ
μ μ 곡νλ€.
2. λλ **κ°μ₯ ν΅μ¬μ μΈ λ¦¬μ€ν¬ λλ λΆνμ€μ±**μ κΈ°λ°ν΄ μ§λ¬Έμ 1κ° λμ§λ€.
3. λ΄κ° λ΅νλ©΄, κ·Έ λ΅μ λ°νμΌλ‘ **꼬리λ₯Ό 무λ λ°©μ**μΌλ‘ μΆκ° μ§λ¬Έμ μ΄μ΄κ°λ€.
4. μ§λ¬Έμ λ€μμ μμμ μμ°μ€λ½κ² 컀λ²νλλ‘ μ λνλ€:
- λ¬Έμ μ μ λ° κ³ κ° νμΈν¬μΈνΈ
- νκ² κ³ κ° μΈκ·Έλ¨ΌνΈ
- μ ν/μλΉμ€μ ν΅μ¬ κ°μΉ
- μμ₯ κ·λͺ¨ λ° κ²½μ νκ²½
- μμ΅ λͺ¨λΈ
- μ€ν μ λ΅ λ° ν ꡬμ±
- μ§ν/μ±κ³΅ κΈ°μ€
## π§ μ§λ¬Έ μ€νμΌ
- μ§λ¬Έμ κ°κ²°νκ³ λμ μ μ΄λ, μ‘΄μ€ μλ ν€μΌλ‘ μμ±ν κ²
- μ: βμ μ§κΈ μ΄ λ¬Έμ κ° μ€μνλ€κ³ μκ°νλμ?β, βλΉμ μ μμ΄λμ΄κ° κΈ°μ‘΄ μ루μ
λ³΄λ€ λμ μ μ 무μμΈκ°μ?β
## β
μΈμ
μ’
λ£ μ‘°κ±΄
- λ΄κ° λ€μ μ€ νλλ₯Ό λͺ
μνλ©΄ μΈμ
μ μ’
λ£νλ€: "λ μ΄μ μ§λ¬Έ μμ", "κ³νμ΄ λͺ
νν΄μ‘μ", λλ "λ".
This method is useful in the early stages of requirements analysis or when improving the quality of detailed requirements. Letβs review the proposal with AI. It may be effective to complete documents together by exchanging questions and answers with AI. In some cases, it is also a good idea to directly input the persona of the interviewer who you want to ask you questions into the AI.
AI only generates the most probable answers based on the given data; it does not have the unique vision or market intuition that a service should have. Services must be released to the market, and it is important to remember that they will be used by actual users.
2. Specification: Draw consistent and detailed blueprints
Once the requirements analysis has determined the direction of what to create, it is necessary to specify the requirements in order to make them concrete. If AI helped sharpen the intuition of those creating services in the requirements phase, it can be used in the specification phase to request consistent production of repetitive tasks or automate certain parts of the work.
Do β : Create drafts of βuser storiesβ and βacceptance criteria.β
User stories and acceptance criteria, which are frequently used in agile development methodologies, are effective methods for describing requirements from the userβs perspective and clarifying completion conditions. The more detailed the requirements for the specifications, the higher the quality of the acceptance criteria.
<ν둬ννΈ>
λλ μ€μ κ²½νμ΄ νλΆν μ μμΌ μ½μΉμ΄μ ν
ν¬λ컬 λΌμ΄ν°λ€.
## π― λͺ©μ
μ£Όμ΄μ§ κΈ°λ₯ λͺ
μΈμ μ μ±
μ λ°νμΌλ‘, κ°λ°νμ΄ λ°λ‘ μμ
ν μ μλλ‘ **Gherkin μ€νμΌμ μ¬μ©μ μ€ν 리**μ **μμ ν μΈμ κΈ°μ€(Acceptance Criteria)**μ μμ±νλΌ.
## π¦ μ
λ ₯ 컨ν
μ€νΈ
- κΈ°λ₯ μ€λͺ
: [μ¬κΈ°μ κΈ°λ₯μ λν μ€λͺ
μ μ
λ ₯νμΈμ. μ: μ¬μ©μλ μ±μμ μνμ μ₯λ°κ΅¬λμ λ΄μ μ μλ€.]
- κ΄λ ¨ μ μ±
λ° μ μ½ μ‘°κ±΄: [μ: λΉνμμ μ΅λ 10κ°κΉμ§λ§ λ΄μ μ μμ. μ¬κ³ κ° μλ μνμ λ΄μ μ μμ λ±]
## π μμ± μ§μ
1. μ¬μ©μ μ€ν 리
- βAs a [μ¬μ©μ μ ν], I want to [νλ], so that [λͺ©ν]β ν¬λ§·μΌλ‘ μμ±ν κ²
- λͺ
ννκ³ λ¨μΌ κΈ°λ₯ λ¨μλ‘ κΈ°μ
2. μΈμ κΈ°μ€ (Acceptance Criteria)
- Gherkin νμ (Given / When / Then)μ μ¬μ©
- λ€μ 3κ°μ§ μλ리μ€λ₯Ό λ°λμ ν¬ν¨ν κ²:
- β
μ μ μλ리μ€: μ‘°κ±΄μ΄ μΆ©μ‘±λ κ²½μ° κΈ°λ₯μ΄ μ λλ‘ μλνλ μν©
- β μ€ν¨ μλ리μ€: μ μ±
λλ μ μ½ μ‘°κ±΄ μλ° μμ λμ λ°©μ
- β μ£μ§ μΌμ΄μ€: κ·Ήλ¨μ /κ²½κ³ μν©μμμ μ²λ¦¬ λ°©μ
3. κ° μΈμ κΈ°μ€μ ν
μ€νΈ κ°λ₯νκ³ , λͺ¨νΈνμ§ μμΌλ©°, ꡬ체μ μΈ μν λ³νλ UI λ°μμ ν¬ν¨ν΄μΌ νλ€.
## π μΆλ ₯ νμ
### μ¬μ©μ μ€ν 리
As a [μ¬μ©μ μ ν], I want to [νλ], so that [λͺ©ν].
---
### μΈμ κΈ°μ€
#### β
μ μ μλ리μ€
- **Given** [μ΄κΈ° 쑰건]
- **When** [νλ λ°μ]
- **Then** [κΈ°λ κ²°κ³Ό]
#### β μ€ν¨ μλ리μ€
- **Given** [...]
- **When** [...]
- **Then** [...]
#### β μ£μ§ μΌμ΄μ€
- **Given** [...]
- **When** [...]
- **Then** [...]
---
## π μμ± μ μ μμ¬ν
- λλ½λ μ μ±
μ΄ μλ€λ©΄ μ§λ¬Ένμ¬ λ³΄μ μμ²
- λ¨μ λ²μμ΄ μλ, μ€μ κ°λ°μκ° λ°λ‘ ꡬν κ°λ₯νλλ‘ κ΅¬μ²΄μ μΌλ‘ μμ±
By submitting the policy document together with the functional specifications, discussions regarding the policy can be conducted prior to the commencement of development. It is also helpful to review the functional specifications one by one. However, if you perform the above prompt for the entire backlog or automate it to generate tasks, you can manage tasks that reflect the requirements specification and policy document.
Do β‘: Maintaining consistency by leveraging existing assets
The more experienced an organization is, or the more experience it has in developing services, the more likely it is to have well-written planning documents and technical specifications. You can create specifications for new services or features by utilizing existing assets. You can either provide existing specifications and use the template, use existing specifications to exchange questions and answers and have the AI organize them, or teach the AI existing templates so that it can suggest better ones.
At this point, it is important to note that the results will be based on the organizationβs key assets, so you must clearly understand how the information you provide will be used before making your request.
Do β’: Automate the βDefinition of Ready (DoR)β checklist
In addition to utilizing prompts, automation is another area where AI excels. After converting requirements into specifications and specifications into tasks with the help of AI, you can set rules for the βDefinition of Readyβ criteria, which determines whether a task is ready to begin development, and manage them according to the business tools you are using. In a situation where AI and human-generated tasks are mixed in the backlog, you can assign the task of reviewing issues and requesting additional information from the appropriate person.
<ν둬ννΈ>
λλ μ°λ¦¬ νμ κΌΌκΌΌν μ€ν¬λΌ λ§μ€ν°(Scrum Master)λ‘ νλνκ³ μλ€.
## π― λͺ©μ
'Definition of Ready(DoR)' 체ν¬λ¦¬μ€νΈλ₯Ό κΈ°μ€μΌλ‘ λ°±λ‘κ·Έ μ΄μλ₯Ό νλμ© κ²ν νκ³ , μμ§ μ€λΉκ° μλ£λμ§ μμ νλͺ©μ΄ μλ€λ©΄ **λ΄λΉμμκ² λͺ
ννκ³ μΉμ ν 보좩 μμ² λ©μμ§λ₯Ό μμ±**νλΌ.
---
## π μ°λ¦¬ νμ Definition of Ready (DoR)
1. β
μ΄μ μ λͺ©μ μ¬μ©μκ° μ΄ν΄ν μ μλ λͺ
νν μΈμ΄λ‘ μμ±λμλκ°?
2. β
μ¬μ©μ μ€ν 리(User Story)λΌλ©΄ Gherkin μμμ μ μ νκ°?
3. β
Task μ΄μλΌλ©΄ 무μμ ν΄μΌ νλμ§ λͺ
νν μ€λͺ
λμλκ°?
4. β
ꡬ체μ μΈ μΈμ κΈ°μ€(Acceptance Criteria)μ΄ μ΅μ 3κ° μ΄μ ν¬ν¨λμλκ°?
5. β
μ μ ν μμ μ΄μκ° λ§ν¬λμλκ°?
6. β
λ΄λΉμκ° μ§μ λμλκ°?
---
## π μμ
μ§μ
1. μ 'κ²ν ν μ΄μ'κ° DoRμ κ° νλͺ©μ μΆ©μ‘±νλμ§ νλͺ©λ³λ‘ νλμ© μ κ²νλΌ.
- νλͺ©λ³ κ²°κ³Όλ λ€μ νμμΌλ‘ μμ±ν κ²:
---
Definition of Ready κ²ν κ²°κ³Ό
β
[μΆ©μ‘± μ¬λΆ] β [κ°λ¨ν μ΄μ ]
β
/ β ... β ...
---
2. νλλΌλ λ―ΈμΆ©μ‘± νλͺ©μ΄ μμ κ²½μ°, λ΄λΉμμκ² λ³΄λΌ **보좩 μμ² λ©μμ§ μ΄μ**μ μμ±νλΌ. λ€μ νμμ λ°λ₯Ό κ²:
---
[λ΄λΉμ μ΄λ¦] μλ
νμΈμπ
ν΄λΉ μ΄μκ° μμ§ μλ νλͺ©μ μΆ©μ‘±νμ§ μμ Definition of Ready μνμ λλ¬νμ§ λͺ»νμ΅λλ€:
[νλͺ© μμ½]
β νμν μ 보: [ꡬ체μ μΌλ‘ μ΄λ€ λ΄μ©μ΄ μ νμνμ§]
μ΄ νλͺ©μ΄ μΆ©μ‘±λλ©΄ λ°λ‘ Sprint Backlogλ‘ μ΄λν μ μμ΅λλ€. μκ³ μ€λ½κ² μ§λ§ μ
λ°μ΄νΈ λΆνλ립λλ€. π
κ°μ¬ν©λλ€!
---
## π‘ μΆκ° 쑰건
- μ¬λ¬ νλͺ©μ΄ λ―ΈμΆ©μ‘±λ κ²½μ°, κ°κ° κ°λ³ νλͺ©μΌλ‘ μ 리
- Taskμ User Storyλ₯Ό ꡬλΆνμ¬ νλ¨
- 보좩 μμ² λ©μμ§ μ΄μμ λ΄λΉμ λ³λ‘ μ μ κ²
- λ무 곡격μ μ΄κ±°λ μ¬λ¬΄μ μΈ ννμ νΌνκ³ , **μ μ€νκ³ νλ ₯μ μΈ ν€**μ μ μ§ν κ²
This automated review system helps ensure that all backlog items maintain a consistent level of quality. The important point to keep in mind is that the recipient of the notification should feel that AI is providing assistance. If you receive too many notifications, you will stop seeing them.
Donβt: Donβt assign vague and abstract tasks as they are.
When assigning work to AI as a business partner, the most important thing to be careful about is assigning ambiguous tasks. Most AI is designed to help users. Therefore, it provides plausible answers to ambiguous questions with a high degree of accuracy. To prevent this situation, you can assign a role to AI so that it can ask questions when it receives ambiguous requests. By utilizing the βmeta-promptβ technique, you can improve the overall quality of collaboration with AI.
<ν둬ννΈ>
# [λ²μ© μμ€ν
ν둬ννΈ] "λͺ
νν ν μλ΅" μμΉ
## μν
λλ μ¬μ©μμ μμ²μ μΆμΈ‘νμ¬ μλ΅μ μμ±νμ§ μκ³ , λ°λμ λͺ
ννκ³ κ΅¬μ²΄μ μΈ μ 보μ κΈ°λ°ν΄μλ§ μλ΅νλ AI μ‘°μμ΄λ€.
## μλ λ°©μ
1. μ¬μ©μμ μμ²μ μμ νλ©΄, μλ νλͺ©λ€μ΄ ν¬ν¨λμ΄ μλμ§ νλ¨νλΌ:
- μμ²μ λͺ
νν λͺ©ν
- μΆ©λΆν λ§₯λ½ λ° λ°°κ²½ μ 보
- ꡬ체μ μΈ μ±κ³΅ κΈ°μ€ λλ μΆλ ₯ νν
2. μ μμ μ€ νλλΌλ λΆμ‘±νλ€κ³ νλ¨λλ©΄, μ λλ‘ μμ²μ λν΄ λ΅λ³μ μμ±νμ§ λ§λΌ.
3. λμ , μ¬μ©μκ° μμ²μ λͺ
νν ν μ μλλ‘ 3~5κ°μ ν΅μ¬ μ§λ¬Έμ 리μ€νΈ ννλ‘ μ μνλΌ.
## μ§λ¬Έ μμ± κΈ°μ€
- κ° μ§λ¬Έμ κ°κ²°νκ³ κ΅¬μ²΄μ μ΄μ΄μΌ νλ€.
- λ€μ μ€ νλ μ΄μμ λͺ©μ μ μΆ©μ‘±ν΄μΌ νλ€:
- μ£Όμ λμ(μ²μ€, μμ μ λ±)μ λͺ
νν ν¨
- μΆλ ₯μ λ²μ, κΈ°κ°, ν¬λ§· λ±μ νΉμ νκ² ν¨
- νμν λ°°κ²½ μ 보λ μ
λ ₯ λ°μ΄ν°λ₯Ό μμ²ν¨
## μμ 1
**μ¬μ©μ μμ²:** "λ§μΌν
λ³΄κ³ μ μ’ μ¨μ€."
**λμ μλ΅:**
"μ νν λ§μΌν
λ³΄κ³ μλ₯Ό μμ±νκΈ° μν΄ λͺ κ°μ§ μ λ³΄κ° λ νμν©λλ€.
1. μ΄ λ³΄κ³ μμ μ£Όμ λ
μλ λꡬμΈκ°μ? (μ: κ²½μμ§, λ§μΌν
ν)
2. μ΄λ€ ν΅μ¬ λ°μ΄ν°λ μ§νκ° ν¬ν¨λμ΄μΌ νλμ? (μ: κ΄κ³ μ±κ³Ό, μ νμ¨)
3. λΆμν κΈ°κ°μ μΈμ μΈκ°μ? (μ: μ§λ λΆκΈ°, μ΅κ·Ό 30μΌ)
4. μ νΈνλ λ³΄κ³ μ ν¬λ§·μ΄ μλμ? (μ: μμ½ μ€μ¬, μκ° μλ£ ν¬ν¨)"
## μμ 2
**μ¬μ©μ μμ²:** "κΈ°νμ μ’ λ§λ€μ΄μ€."
**λμ μλ΅:**
"μ νν κΈ°νμμ λ리기 μν΄ λͺ κ°μ§ μ λ³΄κ° λ νμν©λλ€.
1. κΈ°νμμ μ£Όμ λ λͺ©νλ 무μμΈκ°μ?
2. μ£Όμ λμμ λꡬμΈκ°μ? (μ: ν¬μμ, λ΄λΆ ν)
3. ν¬ν¨λμ΄μΌ ν ν΅μ¬ νλͺ©μ 무μμΈκ°μ? (μ: μμ°, μΌμ , μμ ν¨κ³Ό)
4. μ΄λ€ νμμΌλ‘ μμ±λκΈ°λ₯Ό μνμλμ? (μ: PPT, λ¬Έμ, ν ν¬ν¨)"
μ΄λ¬ν μ§λ¬Έμ ν΅ν΄ μΆ©λΆν μ λ³΄κ° ν보λλ©΄, κ·Έ λ€μμ μμ²λ μμ
μ μννλΌ.
Ask questions proactively and use them wisely
AI can help increase productivity. However, if the direction is unclear, the result will be unclear. The key is to make appropriate requests to AI and receive the desired results, but you should not expect the request and result to magically appear at once. The answer may lie in breaking down big questions into smaller ones, or even discovering the userβs intuition through a completely different persona.
It is important to remember that AI-generated outputs should be viewed as drafts that require review. As time passes and the modelβs performance improves, the quality of the draft may improve, but the responsibility for the output lies with the person. When AI is treated as a partner that helps users focus on valuable tasks, it will become a tool that maximizes potential.