Two Tools That Save Developers' Time: A Well-Written Tech Spec and AI

Kurly's Product Core Frontend Team has established Tech Spec as a core development culture since 2023. Rather than simply documenting features, they use it as a design document that outlines why something is being built and how it will be implemented, turning documentation from a burden into a weapon for collaboration and decision-making.

A Tech Spec consists of four main axes.

• Overview: Organizes both the business background and technical goals together.
• Work Plan: Breaks down implementation units and manages them at the Jira ticket level.
• Implementation: Records core logic, pseudo code, and troubleshooting.
• Project Information: Bundles milestones and related documents for an at-a-glance view.

The effects were clear. PR reviews moved away from a method that took a long time to explain context, toward a method where reviewers understand the background from the document alone and then engage in in-depth discussion. Review time dropped from 1-2 days to around 1-2 hours, and catching issues in advance at the design stage significantly lowered communication costs and rework.

As a real-world case, the product detail page coupon-applied pricing and download project is presented. They faced an issue where the same Coupon_ID was tangled across multi-deal products, causing coupons to be downloaded all at once, but based on the implementation section of the document, they designed a unique key combining Coupon_ID + Deal_Product_No to solve it. They also decided to implement the in-app coupon download feature via WebView instead of native, and in advance checked issues such as response speed to policy changes, reusability, app-web session synchronization, and UI spacing.

Tech Spec also proved powerful in schedule estimation. By breaking tasks down into component and logic units to estimate effort, they could explain "how long it takes" with data rather than intuition, and collaborating departments could immediately understand the complexity.

Now the team is going a step further and experimenting with AI automation. They are building a flow where entering a PRD delivered by a planner generates a Tech Spec draft, and for task decomposition they reference Taskmaster to organize priorities and dependencies. As an example, for a simple to-do/memo page, a structured draft was presented covering layout, date navigation, status cycling, localStorage save/load, and responsive design.

The key point is not that AI writes the document instead of the developer. By shifting to a method where AI lays out the skeleton and the developer fills in business impact, edge cases, and technical details, the burden of writing documentation decreases while design quality increases. Furthermore, these task units become the standard for PRs and reviews, which can raise code review speed and team productivity together.

Going forward, the goal is a flow that creates a Tech Spec from a PRD, links it again with Jira to auto-generate tasks, and extends through to pseudo code. Once a virtuous cycle takes hold—designing with documents first, structuring with AI, and then feeding development and troubleshooting back into the documents—records become not personal notes but a sustainable asset for the team.