The Evolution of AI-Assisted Development: From Prompts to Engineering
As artificial intelligence technology becomes deeply integrated with software development, the developer community is exploring new paradigms that go beyond simple prompt engineering. Recently, the open-source project gstack, led by Y Combinator CEO Garry Tan, has garnered widespread attention, amassing 55.6k stars on GitHub as of March 30, 2026. However, gstack is not an isolated case. obra’s Superpowers and EveryInc’s Compound Engineering are also dedicated to defining AI’s role in software engineering. While all three are AI engineering tools, their core philosophies and implementation paths differ significantly, representing three distinct directions: individual workflows, development discipline, and team collaboration.
Core Philosophies of Three AI Engineering Paths
All three projects aim to build an AI-native software development process, but their distinct focuses make them suitable for different development needs and organizational scales.
gstack: A Full-Lifecycle Workbench for Individual Developers
gstack was created by Garry Tan, leveraging his deep engineering and product experience from Palantir and Posterous. Its core purpose is to be a full-lifecycle AI development workbench for individual developers, covering everything from idea incubation to product launch. It is more than just a collection of code generation tools; it’s a complete workflow that includes stages like office hours (requirements gathering), engineering review, QA (quality assurance), ship (deployment), and retro (retrospective). A standout feature is its integrated persistent Chromium browser service, which can maintain sessions, login states, and cookies, greatly enhancing the efficiency of automated tasks involving web interaction. gstack aims to be an AI co-pilot capable of independently advancing projects, especially suitable for agile development scenarios where product and engineering decisions are tightly intertwined.
Superpowers: A Methodological Framework Emphasizing Development Discipline
The Superpowers project, created by developer obra, finds its core value in introducing strict software engineering discipline to AI agents. It codifies the development task into a clear, stable sequence of steps: brainstorming, planning, TDD (Test-Driven Development), review, and finish. The advantage of this methodology lies in its clear and predictable process, which helps developers enforce mature engineering practices (like test-first development and code review) on the AI’s output. Superpowers is more like a lightweight process engine than an all-encompassing toolset. It focuses on regulating the AI’s behavior, making it perform more like a well-trained, rule-abiding engineer. It offers an ideal starting point for developers who want to integrate structured AI capabilities into their existing development habits.
Compound Engineering: A System for Team Collaboration and Knowledge Persistence
Compound Engineering, launched by EveryInc, shifts the focus to team-level collaboration and knowledge management. Its design emphasizes the division of labor among multiple roles and the continuous reuse of experience. The system includes various specialized agents, such as reviewers and researchers with different perspectives, and provides commands like compound and compound-refresh. These are used to distill successful interaction experiences and code patterns into a knowledge base that can be called upon for future tasks. This design makes it more like an enterprise-grade engineering collaboration system than a personal tool. Its complexity is higher, but for teams that require large-scale collaboration and value code standards and knowledge inheritance, its long-term value is more pronounced.
Methodology, Tooling, and Scenarios: A Comparative Analysis
Viewed from different dimensions, the differences between the three become clearer, forming a multi-layered landscape of AI engineering solutions.
Development Methodology: Superpowers is the most focused, concentrating on process discipline. gstack is the most complete, covering the end-to-end individual development lifecycle. Compound Engineering is the most team-oriented, emphasizing role-based division of labor and knowledge accumulation.
Degree of Tooling: gstack is the most productized, offering a nearly out-of-the-box ‘personal AI development environment’. Compound Engineering is a multi-component, extensible system that relies on external tools like Context7 and agent-browser. Superpowers is relatively restrained, prioritizing the process itself over specific tool implementations.
- Use Cases and Users: Indie developers or small agile teams can gain rapid benefits from gstack. Developers looking to integrate AI into a standardized engineering workflow will find Superpowers an ideal choice for building good habits. When a team scales and faces challenges with collaboration efficiency, code quality, and experience reuse, the systematic advantages of Compound Engineering will become evident.
Conclusion: How to Choose the Right AI Engineering Path for You
Overall, these three projects are not simple competitors. Instead, they represent solutions for AI-assisted development at different maturity stages. They answer three progressively layered questions:
- Superpowers: How should an AI agent work in a disciplined manner?
- gstack: How can an AI agent participate completely in a project’s development process?
- Compound Engineering: How should a team organize and reuse its AI engineering capabilities?
For developers or teams considering adopting these tools, the first task is to clarify their core needs. If it’s for a personal project or exploratory development, starting with gstack or Superpowers would be a highly efficient choice. If a team is already facing collaboration bottlenecks and wishes to build a scalable, reusable AI engineering system, then investing time to research and deploy Compound Engineering would be a more strategic decision.
