Epistemically-Driven Agentic Software Engineering System. EDASES is a research programme investigating how non-programmers can safely and effectively use heterogeneous AI systems to engineer software. The project explores how software engineering can move from conversational interaction with individual AI models to structured, evidence-driven collaboration between humans, AI systems and automated tooling. The long-term objective is to develop a methodology that is:

• verifiably safe at every stage
• evidence-driven rather than opinion-driven
• independent of any individual AI model or provider
• mechanically enforceable wherever practical
• capable of evolving through empirical research

The project consists of three closely related layers.

EDASES is the research programme. Its purpose is to investigate, evaluate and validate improved approaches to AI-assisted software engineering through experimentation and evidence. The outputs of EDASES are research findings rather than software.

The Agentic Software Engineering System (ASES) is the methodology produced by the EDASES research programme. ASES defines how AI-assisted software engineering should be conducted. It is intentionally independent of any particular implementation, AI provider or software platform.

The execution engine is a software implementation of the ASES methodology. Its purpose is not to replace software engineers, but to execute and enforce the methodology mechanically by:

• maintaining engineering state
• preserving reasoning and evidence
• coordinating heterogeneous AI capabilities
• reducing predictable human and AI error
• supporting efficient software engineering workflows

The execution engine is an implementation of ASES, not the methodology itself.

Current research investigates questions including:

• How can reasoning remain traceable throughout software engineering?
• How should epistemic relationships be represented and preserved?
• How can AI capabilities be evaluated objectively?
• How should multiple AI systems collaborate effectively?
• How can software engineering methodology be executed mechanically rather than procedurally?

This repository contains research, methodology and implementation planning. Canonical documentation is organised by abstraction level rather than implementation status.

New contributors should begin with:

1. ORIENTATION.md
2. docs/standards/Documentation Standard.md
3. The foundational documents referenced from the orientation guide

AI contributors should additionally read:

• AGENTS.md

The project is currently transitioning from exploratory research toward the design of a methodology execution engine.

Recent work has established:

• a separation between research (EDASES), methodology (ASES) and implementation
• reasoning as the primary object of interest rather than source code history
• epistemic relationships as the foundation of engineering knowledge
• levels of abstraction as the organising structure of the project
• methodology execution as the next stage of research

Contributions should follow the project's documentation standards and methodology.

Please read ORIENTATION.md before making architectural, methodological or implementation changes.

AI agents should additionally follow the instructions contained within AGENTS.md.

See the accompanying LICENSE file for licensing information.

Project documentation follows a common documentation standard that defines document classification, abstraction layers, dependencies and relationships.

Contributors should begin with:

1. ORIENTATION.md
2. docs/standards/Documentation Standard.md

The orientation guide introduces the project structure, while the documentation standard defines how canonical project knowledge is represented.