Structured-Prompt-Driven Development (SPDD) is an innovative workflow developed by Thoughtworks’ internal IT team to harness the power of LLM programming assistants in a team setting. Unlike individual use, SPDD treats prompts as first-class artifacts—versioned alongside code—to ensure alignment with business goals. This approach requires developers to master three core skills: alignment, abstraction-first thinking, and iterative review. The following Q&A unpacks how SPDD works and why it’s gaining traction.
What exactly is Structured-Prompt-Driven Development?
SPDD is a method that integrates prompt engineering into the software development lifecycle. Instead of viewing prompts as disposable instructions to an AI assistant, SPDD treats them as formal, version-controlled artifacts. This means prompts are written, reviewed, and maintained just like source code. The workflow typically involves defining a high-level prompt that captures business requirements, then refining it through abstraction and iterative feedback. The goal is to keep the AI-generated code aligned with project objectives across the entire team, not just for individual developers.
Why treat prompts as first-class artifacts in version control?
Treating prompts as first-class artifacts ensures that every team member can see exactly what instructions were given to the LLM at each stage. This transparency reduces misunderstandings and makes it easy to roll back or compare changes. It also creates a clear audit trail: if a generated feature isn’t meeting business needs, you can trace the issue back to a specific prompt. By keeping prompts in the same repository as code, teams establish a single source of truth. This practice fosters collaboration, as developers can discuss and improve prompts just like pull requests.
What three key skills do developers need for effective SPDD?
- Alignment – Ensuring the prompt accurately reflects the business requirement and intended outcome.
- Abstraction-first – Starting with a high-level, conceptual prompt before diving into details, much like designing software architecture.
- Iterative review – Regularly revisiting and refining prompts based on feedback and test results, similar to agile development cycles.
These skills shift the developer’s role from writing every line of code to orchestrating the AI through well-crafted prompts.
How does SPDD align development with business needs?
Alignment is achieved by making the prompt the primary vehicle for expressing business intent. A well-structured prompt states the problem, constraints, and expected behavior in plain language. Because the prompt is versioned and shared, product owners and stakeholders can review it before any code is generated. This early feedback loop catches misinterpretations before they become costly rewrites. Additionally, because prompts evolve with the project, they naturally track shifting priorities—keeping the generated code continuously in sync with business goals.
How does the abstraction-first approach improve outcomes?
The abstraction-first approach encourages developers to think conceptually before writing details. Instead of telling the AI to “create a button that turns blue on click,” they first define the broader goal: “implement a toggle control that reflects user preference.” This high-level prompt forces team members to agree on intent and scope. Once the abstraction is stable, subsequent prompts can add specifics like styling or event handling. This method reduces the risk of the AI producing code that meets literal instructions but misses the bigger picture.
What role does iterative review play in SPDD?
Iterative review is the engine that refines prompts over time. After an initial prompt generates code, the team tests it against acceptance criteria. Any gaps or errors are documented, and the prompt is updated accordingly—often with more precise language or additional context. This cycle repeats until the output meets quality standards. Because prompts are versioned, the team can see the evolution of instructions and learn which phrasings work best. Iterative review also catches regressions when new prompts overwrite older ones.
Can you walk through a simple example from the GitHub repository?
Wei Zhang and Jessie Jie Xia provide a concrete SPDD example on GitHub. The example focuses on building a simple to-do list application. The team starts with a high-level prompt describing the core feature: “Create a to-do list that allows adding, completing, and deleting tasks.” They then break this down into sub-prompts for each operation. Each sub-prompt is reviewed iteratively, and the generated code is tested for correctness. The repository shows how prompts are organized in markdown files alongside the source code, illustrating the first-class artifact principle.