Thoughtworks Introduces SPDD Methodology, Bringing AI Programming From Individual Use to Team Collaboration

A New Solution to the Team-Level Challenge of LLM Coding Assistants

Large language model (LLM) coding assistants have already demonstrated significant value at the individual developer level. Tools like GitHub Copilot and Cursor are profoundly changing how developers write code on a daily basis. However, a long-overlooked question remains: when LLM coding assistants expand from being a personal tool to a team collaboration scenario, how can teams ensure the quality, consistency, and reusability of prompts?

The internal IT team at Thoughtworks, a globally renowned technology consultancy, has drawn on real-world project experience to propose a methodology and workflow called Structured-Prompt-Driven Development (SPDD), offering a systematic solution to this challenge.

What Is SPDD?

The core philosophy of SPDD can be summarized in one sentence: Treat prompts as first-class artifacts on par with source code.

In traditional AI-assisted development workflows, interactions between developers and LLMs tend to be impromptu and disposable — a prompt is written, the generated output is collected, and the prompt itself is immediately discarded. While this approach is acceptable for individual development, it creates numerous problems in team settings:

• Knowledge cannot be accumulated: Excellent prompting techniques remain locked in individual minds and cannot be shared across the team
• Quality is difficult to ensure: Prompts written by different team members vary widely in quality, leading to inconsistent outputs
• Processes are untraceable: It is impossible to trace back which prompt was used to generate a particular piece of code
• Collaboration lacks standards: Team members work in silos without a unified workflow

The SPDD methodology is designed specifically to address these pain points. It requires teams to save carefully crafted structured prompts alongside code in version control systems, turning prompts into trackable, reviewable, and iterable development assets.

SPDD Workflow in Detail

Wei Zhang and Jessie Jie Xia from Thoughtworks published detailed documentation and examples of the SPDD workflow on GitHub. Based on publicly available materials, the SPDD workflow comprises the following key elements:

1. Structured Prompt Design

SPDD emphasizes that prompts should not be casually written fragments of natural language but should follow a unified structured template. This structured design ensures that every prompt includes clear contextual information, task descriptions, constraints, and expected output formats, thereby improving the stability and quality of LLM-generated results.

2. Prompts Under Version Control

This is one of SPDD's most innovative practices. Prompt files are stored in Git repositories alongside source code, receiving the same version management treatment — with full change history, code reviews, branching, and merge requests. This means teams can clearly trace the motivation behind every prompt modification and its resulting impact.

3. Team Collaboration Workflow

Under the SPDD framework, team members can reuse and improve upon an existing prompt library, avoiding redundant effort. New team members can also quickly learn the team's best practices by studying the high-quality prompts accumulated in the repository.

Why SPDD Deserves Attention

From a broader perspective, the emergence of SPDD reflects an important shift underway in AI-assisted development: the evolution from a personal productivity tool to a team engineering practice.

Most industry discussions about LLM coding assistants currently focus on individual productivity gains — how much faster a developer can write code with AI, or how much time is saved. But in real enterprise development environments, software engineering has always been a team activity. How to maximize the value of AI coding assistants at the team level is a question that urgently needs answering.

As a company with deep expertise in engineering practices such as agile development and continuous integration, the fact that Thoughtworks' own internal team developed the SPDD methodology lends it considerable practical credibility. This is not an academic concept drafted on paper but a distillation of team experience validated through real projects.

Furthermore, the idea of treating prompts as first-class artifacts is philosophically aligned with the concept of Infrastructure as Code (IaC) in software engineering. Just as IaC brought infrastructure configurations under version control to enable traceable and reproducible operations practices, SPDD brings prompts under version control with the potential to enable traceable and reproducible AI-assisted development practices.

Industry Trends and Future Outlook

As AI coding assistants continue to penetrate enterprises at increasing rates, team-level methodologies like SPDD will become increasingly important. Foreseeable trends include:

Professionalization of prompt engineering: Prompts will no longer be a developer's byproduct but engineering assets that require dedicated design, testing, and maintenance. Specialized prompt quality assessment tools and CI/CD integration solutions may emerge in the future.

Reshaping of development workflows: The traditional "requirements → design → coding → testing" process may evolve into a new paradigm of "requirements → prompt design → AI generation → human review → iterative optimization," and SPDD represents an early exploration in this direction.

Community-driven best practices: As more teams adopt similar approaches, high-quality structured prompt templates and workflow standards are expected to be shared and refined within open-source communities.

For technical managers and development teams exploring how to systematically use LLM coding assistants at the team level, Thoughtworks' SPDD methodology offers a worthy starting point. Interested readers can visit GitHub to review the detailed examples and documentation published by Wei Zhang and Jessie Jie Xia for specific implementation details of this workflow.