Introduction to Structured Prompt-Driven Development
Large language model (LLM) coding assistants have proven tremendously useful for individual developers, but scaling their benefits to entire teams requires a more structured approach. Thoughtworks’ internal IT organization has pioneered a methodology called Structured Prompt-Driven Development (SPDD) that transforms how teams collaborate with AI. Instead of treating prompts as ephemeral inputs, SPDD makes them a first-class artifact—stored in version control alongside code, aligned with business needs, and continuously refined through a defined workflow.
The SPDD Workflow in Practice
Wei Zhang and Jessie Jie Xia provide a concrete example of the SPDD workflow on GitHub. The process revolves around three core phases:
1. Prompt Design and Alignment
Before writing any code, developers collaboratively design prompts that capture business requirements, acceptance criteria, and technical constraints. These prompts are written in a domain-specific language or structured template that the team maintains. By treating prompts as specifications, the team ensures that every AI-generated output directly serves a business goal.
2. Abstraction-First Development
Instead of asking the AI to generate entire functions or modules, SPDD encourages an abstraction-first mindset. Developers define interfaces, data structures, and high-level logic first, then use prompts to fill in implementation details. This separates what needs to be done from how it is done, making the code easier to review and maintain.
3. Iterative Review and Refinement
Each prompt-generated code snippet is reviewed against the original requirements, and the prompt itself is refined based on what the AI produces. The team creates a feedback loop where imperfect outputs lead to better prompts, which in turn produce higher-quality code. All prompt versions are tracked in version control alongside code changes.
Three Essential Skills for Developers
Thoughtworks’ experience reveals that developers need to master three key skills to be effective in SPDD:
- Alignment – the ability to translate business needs into precise, testable prompts that leave no room for ambiguity.
- Abstraction-first – thinking in terms of APIs, contracts, and modularity before diving into implementation.
- Iterative review – evaluating AI outputs critically and refining prompts systematically over multiple cycles.
These skills shift the developer’s role from code producer to prompt engineer and quality gate, a change that teams must embrace for SPDD to succeed.
Treating Prompts as First-Class Artifacts
The most distinctive feature of SPDD is that prompts are version-controlled alongside code. This practice brings several benefits:
- Traceability: Every line of code can be traced back to the prompt that produced it, making audits and debugging easier.
- Reproducibility: Teams can replay old prompts to understand why code was written a certain way.
- Collaboration: Prompts become a shared language between developers, product owners, and other stakeholders, reducing miscommunication.
- Continuous improvement: As prompts evolve, they encode institutional knowledge about the project.
Implementing SPDD in Your Team
To adopt SPDD, start small. Pick a single feature or module and have the team write structured prompts for it. Use a template that includes:
- Role context (e.g., “You are a senior Python developer”)
- Business requirement
- Acceptance criteria
- Constraints (e.g., performance, security)
- Expected output format
Store these in a /prompts directory in your repository. After each sprint, review which prompts produced high-quality code and which need improvement. Over time, the team will build a library of proven prompts that accelerate development.
For a working example, refer to the SPDD demo repository by Zhang and Xia.
Conclusion: Beyond Individual Productivity
SPDD moves LLM usage from a personal productivity hack to a disciplined, team-wide engineering practice. By making prompts a central, version-controlled artifact, teams gain alignment with business goals, improve code quality, and build a reusable knowledge base. The three skills of alignment, abstraction-first, and iterative review are the building blocks of this transformation.
As AI coding tools become ubiquitous, structured approaches like SPDD will differentiate teams that simply use AI from those that master it for consistent, high-value outcomes.