TLDR: Agentsway is a novel software development methodology designed for teams that integrate autonomous AI agents as first-class collaborators. It addresses the limitations of traditional human-centric methodologies by introducing a structured lifecycle with human orchestration and specialized AI agents for planning, prompting, coding, testing, and LLM fine-tuning. The framework emphasizes governance, privacy-by-design, and responsible AI principles, enabling continuous learning and iterative improvement. A real-world use case in legal case handling automation demonstrated its effectiveness in creating coherent plans and accurate prompts.
The landscape of software development is undergoing a profound transformation with the rise of Agentic AI. Traditionally, methodologies like Agile and Kanban were designed for human-centric teams, but these approaches are proving insufficient in environments where autonomous AI agents actively participate in planning, coding, testing, and continuous learning.
Addressing this critical gap, a new software development framework called “Agentsway” has been introduced. This innovative methodology is specifically crafted for ecosystems where AI agents are considered first-class collaborators, working alongside human teams.
At its core, Agentsway establishes a structured lifecycle that emphasizes human orchestration, robust governance, and privacy-preserving collaboration among a team of specialized AI agents. The framework defines distinct roles for various agents, each contributing to an iterative improvement and adaptive learning process throughout the development cycle.
The Agentsway Team: Human and AI Collaborators
In the Agentsway model, the human plays a crucial role as the orchestrator. This involves interpreting high-level business goals, engaging with stakeholders, and ensuring that the final software aligns with organizational objectives. The human also validates artifacts generated by the AI agents and provides essential governance and ethical oversight.
Surrounding the human orchestrator is a suite of specialized AI agents:
-
Planning Agent: This agent acts as the central reasoning component. It analyzes project documents and requirements, then breaks them down into executable tasks. Leveraging fine-tuned Large Language Models (LLMs), it generates detailed task descriptions, resource estimates, and project pitches, which are then reviewed and approved by the human orchestrator.
-
Prompting Agent: Bridging planning and implementation, the Prompting Agent constructs detailed, context-aware prompts specifically tailored for the Coding Agents. These prompts incorporate functional requirements, coding style, and integration dependencies, ensuring clarity and consistency.
-
Coding Agents: These agents are responsible for translating approved prompts into executable code. They operate within defined project environments, adhering to organizational coding standards and architectural constraints, and autonomously implement features and refactor modules.
-
Testing Agents: Ensuring quality and reliability, Testing Agents perform automated unit, integration, and regression tests. They also conduct static analysis and vulnerability scans, producing reports that guide corrective actions for both humans and coding agents.
-
Fine-Tuning Agents: These agents form the learning and improvement layer. After each development cycle, they collect data—such as prompts, generated code, and testing feedback—to incrementally refine pre-trained LLMs. This retrospective process enhances contextual accuracy and adaptability over time, all within secure, privacy-preserving environments.
A key aspect of Agentsway is its commitment to responsible AI principles. This is achieved through the integration of multiple fine-tuned LLMs, which operate as a consortium, coupled with a dedicated reasoning LLM. This ensemble-based approach ensures balanced, multi-perspective reasoning and enhances the accuracy and accountability of agentic decisions.
Also Read:
- TDFlow: Enhancing Software Development with Test-Driven AI Workflows
- Structured Code Generation with LLMs: A Lifecycle Approach
Real-World Application: Legal Case Handling
To demonstrate its practical applicability, Agentsway was applied to a use case involving legal case handling automation. The objective was to automate document retrieval, case summarization, and legal question answering across large legal corpora, while maintaining human oversight and data privacy. The framework successfully enabled agents to define objectives, decompose requirements, and execute workflows for tasks like context-aware Q&A and case summarization.
The evaluation of Agentsway’s components—the Planning Agent, Prompting Agent, and the Fine-Tuning process—showed promising results in generating coherent plans, accurate prompts, and demonstrable improvements through iterative fine-tuning. This indicates that Agentsway can reliably transform abstract development goals into structured, executable workflows.
Agentsway represents a significant advancement in software engineering, offering a foundational step toward the next generation of AI-native, self-improving software development methodologies. It is the first research effort to introduce a dedicated methodology explicitly designed for AI agent-based software engineering teams. For more in-depth information, you can read the full research paper here: Agentsway Research Paper.
