My PRD Became My API Design - Thanks to AI
- Tassio Batista
- Mar 5
- 2 min read
Updated: Mar 11
A few years ago, at a previous job as a Technical Product Manager, designing an API sparked endless debates.
How should the YAML look?
Which error codes should we return?
Should this be a POST or a PUT?
As a PM, I tried to stay out of technical prescription. My role was to define the business rules and ensure the API served a real need, not dictate how it needed to be built. But watching the team getting stuck in the nitty-gritty of spec details made me wonder:
What if we could skip the back-and-forth and move faster?
So, I started experimenting.
Can My PRD generate an API Design?
With AI, I wondered: Could a well-structured PRD be enough to generate a first draft of an API design? The goal was to avoid prescriptive details (no enforcing specific endpoints, status codes, or architecture). Just a clear business use case and its logic.
So, I put AI to the test.
After refining my prompts, I reached a point where my PRD was solid enough to create a first version of the yaml that would skip some meetings we used to have and would let the team focus on what truly mattered. Instead of debating request formats, we would spend time refining business use cases and outcomes.
My thoughts
Although AI has been extremely helpful for this use case, I don’t see it as a silver bullet that can replace the team's expertise. For now, it’s an effective tool for eliminating busywork - handling tedious setup so we can focus on what truly matters:
Identifying risks and gaps in the overall solution
Validating the solution with customers
Accelerating delivery without sacrificing quality
While drafting the PRD, I intentionally avoided defining requirements tied to a specific API architecture style. As a PM, understanding design standards across different architectures (e.g., REST, GraphQL) is valuable, but staying at a higher level gives engineering the flexibility to explore creative solutions that best address the problem.
How about you? Have you used AI to speed up API design? What worked (or didn’t) for you?
Comments