Structured-Prompt-Driven Development (SPDD) is a workflow developed by Thoughtworks' internal IT organization to help teams use large language model (LLM) programming assistants effectively. Instead of treating prompts as one-off interactions, SPDD treats them as first-class artifacts, stored in version control alongside code, to ensure consistency, alignment with business needs, and continuous improvement. This approach shifts the focus from individual developer productivity to team-wide collaboration, requiring three key skills: alignment, abstraction-first, and iterative review. Below, we answer common questions about this methodology.
What is Structured-Prompt-Driven Development (SPDD)?
Structured-Prompt-Driven Development (SPDD) is a method for integrating LLM programming assistants into team-based software development workflows. Unlike ad-hoc prompt usage by individual developers, SPDD treats prompts as structured, reusable artifacts that are version-controlled alongside the codebase. This ensures that every prompt is deliberate, traceable, and aligned with project goals. Thoughtworks' IT organization pioneered SPDD to address challenges like inconsistent code generation, lack of business alignment, and knowledge silos. By treating prompts as code, teams can review, refine, and share them, creating a shared repository of effective patterns. SPDD also emphasizes an abstraction-first approach, where developers design high-level logic before diving into implementation details, and iterative review to catch errors early. The result is a more reliable, collaborative, and business-focused use of AI coding assistants.
Why treat prompts as first-class artifacts?
Treating prompts as first-class artifacts means they are managed with the same rigor as source code—stored in version control, subject to reviews, and tied to specific project versions. This practice prevents prompt-related issues such as drift (where prompts become outdated as code changes), inconsistency (different developers using different prompts for similar tasks), and opacity (lack of understanding of how code was generated). By keeping prompts under version control, teams can track changes, revert to working versions, and audit how AI contributed to the codebase. This approach also facilitates reuse: a validated prompt for generating a REST endpoint can be shared across the team. Moreover, it aligns development with business needs because prompts are explicitly crafted to reflect requirements, ensuring the LLM produces code that matches the intended functionality.
What are the three key skills developers need for SPDD?
Thoughtworks identifies three essential skills for effective SPDD: alignment, abstraction-first, and iterative review. Alignment means ensuring that prompts directly map to business requirements and technical context—avoiding generic instructions that yield irrelevant code. Abstraction-first encourages developers to define the high-level structure (interfaces, data flows, core logic) before generating low-level implementation, allowing the LLM to fill in details within a clear framework. Iterative review involves constantly checking generated code for correctness, security, and adherence to standards, and refining prompts based on feedback. These skills help developers move beyond passive acceptance of AI output toward active, thoughtful collaboration with the tool, resulting in higher-quality, maintainable code.
How does SPDD align development with business needs?
SPDD aligns development with business needs by embedding requirements directly into prompts. Instead of relying on a developer's memory or external documents, SPDD prompts are crafted to include explicit business rules, user stories, or acceptance criteria. For example, a prompt for generating a discount calculation function might include the business logic: "Apply 10% off for orders over $100, but only for new customers." These prompts are stored with the code, making business intent transparent and reviewable. As requirements evolve, prompts are updated in version control, ensuring the LLM consistently generates code that reflects current needs. This traceability also helps stakeholders verify that the implementation matches expectations, reducing miscommunication and rework. By keeping prompts aligned with business objectives, SPDD turns AI assistants into reliable partners in building value-driven software.
How does SPDD differ from using LLMs individually?
Individual developers often use LLM assistants in an ad-hoc manner—typing a prompt, getting code, and moving on. This approach can be fast for simple tasks but creates problems in team settings: prompts are lost, results vary, and there is no shared understanding of how code was generated. SPDD transforms this into a structured, repeatable process. It mandates that prompts be documented, versioned, and reviewed, turning a personal tool into a team asset. For example, when a developer writes a prompt to generate a data model, they commit that prompt to the repository. Other team members can examine it, suggest improvements, or reuse it. Over time, the team builds a library of proven prompts, reducing duplicate work and ensuring consistency. This collaborative approach is especially valuable for complex projects requiring coordination across multiple developers and business units.
Can you give a simple example of SPDD in action?
Wei Zhang and Jessie Jie Xia provide a concrete example on GitHub (see their repository for details). Suppose a team needs to create a user registration API. In SPDD, they first write a prompt that specifies the endpoint path, HTTP method, required fields (username, email, password), validation rules (e.g., email format, password length), and business logic (e.g., send confirmation email). The prompt is saved as a text file in the code repository. The developer runs the LLM with this prompt to generate the code. After generation, the team reviews the code against the prompt and business requirements, making adjustments if needed. The prompt itself is also reviewed for clarity and completeness. Once approved, both prompt and code are committed. Later, if the registration logic changes (e.g., add phone number field), the prompt is updated first, then the code is regenerated or edited accordingly. This keeps the entire team synchronized and the AI output predictable.
What benefits has Thoughtworks seen from SPDD?
Thoughtworks' internal IT organization reports several benefits from adopting SPDD. First, code quality improves because prompts are carefully crafted and reviewed, leading to more accurate and secure outputs. Second, team productivity increases as reusable prompts eliminate redundant prompt-writing and reduce debugging time. Third, business alignment strengthens: prompts explicitly encode requirements, so generated code consistently reflects stakeholder intent. Fourth, knowledge transfer becomes easier—new team members can learn from the prompt repository what the AI is expected to produce and why. Finally, governance improves because every code generation is traceable to a specific prompt and version. These benefits make SPDD an appealing workflow for any team that relies heavily on LLM assistants and wants to move beyond individual tinkering toward professional, collective AI-assisted development.