Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Compiler.next: A Search-Based Compiler to Power the AI-Native Future of Software Engineering

Filipe R. Cogo, Gustavo A. Oliva, Ahmed E. Hassan

TL;DR

Compiler.next introduces a search-based compiler designed to transform human intents into AI-native FMware (Promptware/Agentware) by jointly optimizing prompts, configurations, and coordination architectures. It addresses the need for dynamic, multi-objective optimization across accuracy, latency, and cost in SE3.0, and presents a concrete architecture with semantic caching, scenario expansion, and distributed synthesis to enable scalable, reproducible FMware compilation. The paper provides an initial validation on HumanEval-Plus and outlines a comprehensive R&D roadmap with ten Calls for Action to advance end-to-end FMware optimization, interoperability, reproducibility, and community data sharing. This framework aims to democratize software development by enabling intent-driven, auto-synthesized AI-powered systems, while aligning with established SE practices such as MBSE, SPLs, and self-adaptive systems. The practical impact lies in enabling faster, cost-aware, and more reliable generation of adaptive AI-native software across domains, fostering a foundation for SE3.0 tooling and workflows.

Abstract

The rapid advancement of AI-assisted software engineering has brought transformative potential to the field of software engineering, but existing tools and paradigms remain limited by cognitive overload, inefficient tool integration, and the narrow capabilities of AI copilots. In response, we propose Compiler.next, a novel search-based compiler designed to enable the seamless evolution of AI-native software systems as part of the emerging Software Engineering 3.0 era. Unlike traditional static compilers, Compiler.next takes human-written intents and automatically generates working software by searching for an optimal solution. This process involves dynamic optimization of cognitive architectures and their constituents (e.g., prompts, foundation model configurations, and system parameters) while finding the optimal trade-off between several objectives, such as accuracy, cost, and latency. This paper outlines the architecture of Compiler.next and positions it as a cornerstone in democratizing software development by lowering the technical barrier for non-experts, enabling scalable, adaptable, and reliable AI-powered software. We present a roadmap to address the core challenges in intent compilation, including developing quality programming constructs, effective search heuristics, reproducibility, and interoperability between compilers. Our vision lays the groundwork for fully automated, search-driven software development, fostering faster innovation and more efficient AI-driven systems.

Compiler.next: A Search-Based Compiler to Power the AI-Native Future of Software Engineering

TL;DR

Compiler.next introduces a search-based compiler designed to transform human intents into AI-native FMware (Promptware/Agentware) by jointly optimizing prompts, configurations, and coordination architectures. It addresses the need for dynamic, multi-objective optimization across accuracy, latency, and cost in SE3.0, and presents a concrete architecture with semantic caching, scenario expansion, and distributed synthesis to enable scalable, reproducible FMware compilation. The paper provides an initial validation on HumanEval-Plus and outlines a comprehensive R&D roadmap with ten Calls for Action to advance end-to-end FMware optimization, interoperability, reproducibility, and community data sharing. This framework aims to democratize software development by enabling intent-driven, auto-synthesized AI-powered systems, while aligning with established SE practices such as MBSE, SPLs, and self-adaptive systems. The practical impact lies in enabling faster, cost-aware, and more reliable generation of adaptive AI-native software across domains, fostering a foundation for SE3.0 tooling and workflows.

Abstract

The rapid advancement of AI-assisted software engineering has brought transformative potential to the field of software engineering, but existing tools and paradigms remain limited by cognitive overload, inefficient tool integration, and the narrow capabilities of AI copilots. In response, we propose Compiler.next, a novel search-based compiler designed to enable the seamless evolution of AI-native software systems as part of the emerging Software Engineering 3.0 era. Unlike traditional static compilers, Compiler.next takes human-written intents and automatically generates working software by searching for an optimal solution. This process involves dynamic optimization of cognitive architectures and their constituents (e.g., prompts, foundation model configurations, and system parameters) while finding the optimal trade-off between several objectives, such as accuracy, cost, and latency. This paper outlines the architecture of Compiler.next and positions it as a cornerstone in democratizing software development by lowering the technical barrier for non-experts, enabling scalable, adaptable, and reliable AI-powered software. We present a roadmap to address the core challenges in intent compilation, including developing quality programming constructs, effective search heuristics, reproducibility, and interoperability between compilers. Our vision lays the groundwork for fully automated, search-driven software development, fostering faster innovation and more efficient AI-driven systems.

Paper Structure

This paper contains 33 sections, 5 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Background and related work
  3. Traditional compilers
  4. Deep learning compilers
  5. Prompt compilers
  6. Synthesizing Intents into FMware
  7. FMware and related concepts
  8. FMware and cognitive architectures
  9. The role of compilers in the se30 era
  10. The challenging task of synthesizing intents into FMware
  11. State of the practice
  12. Positioning SE 3.0 Compilers within Broader SE Paradigms
  13. Compiler.next
  14. The search mechanism
  15. Illustrative Example: Optimizing Prompts for Code Generation
  16. ...and 18 more sections

Figures (5)

  • Figure 1: Example of an FMware module with one FMware component comprised of a Promptware and an Agentware components.
  • Figure 2: RAG pipeline (adapted from OPEA's reference RAG pipeline opea2024)
  • Figure 3: The technology stack of Compiler.next.
  • Figure 4: Search iteration steps in Compiler.next.
  • Figure 5: The operation of one of Compiler.next's optimizers in our running example of an FMware for code generation. The dashed box indicates optional steps.