AutoDev: Building an Autonomous AI Software Engineer for Live Project Environments

The problem

As Omdena scales increasingly complex, multi-repository and multi-team AI projects, a substantial portion of engineering effort is spent on coordination, retrieving project context, and handling repetitive operational tasks rather than solving core technical problems.

Traditional AI assistants operate as passive chat interfaces. They do not actively participate in engineering systems, nor can they meaningfully interact with real project environments, tickets, repositories, or production interfaces.

Recent advances in agentic AI systems introduce the possibility of action-oriented agents that operate inside real software workflows. However, it remains unclear whether an autonomous AI agent can function as a true project participant within a live engineering environment.

There is currently no evidence-based validation of whether such an agent can:

• Log into and operate within a live project.
• Understand tickets and project metadata.
• Clone and interpret repository structures.
• Interpret UI elements in production systems.
• Respond accurately within real engineering workflows.

Without structured experimentation, the feasibility and limitations of an Autonomous Software Engineer remain speculative.

The project goals

This Innovation Challenge proposes building an OpenClaw-powered Autonomous Engineering Agent Prototype embedded directly inside a real Umaku.ai project.

The objective is experimental validation, not production automation.

The solution includes:

• Deploying an OpenClaw-powered agent in a dedicated AWS EC2 environment
• Enabling persistent runtime with memory
• Integrating with Umaku.ai for ticket assignment and notification handling
• Enabling read-only GitHub repository cloning
• Building structured repository interpretation capabilities
• Assembling context from:
  • Tickets
  • Comments
  • Repository content
  • Project metadata
• Posting contextual responses directly inside Umaku.ai
• Experimentally interpreting UI elements such as:
  • Project structure
  • Ticket metadata (status, assignees, priority)
  • Navigation and layout
• Logging ambiguities, failures, and misinterpretations
• Generating observability artifacts including behavioral traces and uncertainty tracking

The outcome will be a research-grade prototype capable of participating in live engineering workflows under controlled constraints.

Impact of the Problem

If validated, this experiment could redefine how AI agents participate in software development environments.

Engineering Teams

• Reduced manual effort in context retrieval.
• Assistance in navigating complex repositories.
• Faster clarification of ticket-related questions.

Product Owners

• Improved visibility into contextual reasoning within engineering workflows.
• Enhanced support for ticket interpretation and metadata analysis.

Multi-Team AI Projects

• Reduced coordination overhead.
• Potential improvement in response time to engineering queries.
• Structured logging of ambiguities and failure patterns.

Organizational Impact

• Evidence-based insight into the feasibility of Autonomous Software Engineers.
• Clear understanding of capability boundaries.
• Architectural guidance for future evolution.

The Goals

This project aims to experimentally validate whether an autonomous AI agent can function as an active participant inside a real Omdena software project within 8 weeks.

The agent must demonstrate the ability to:

• Log into and participate in a real Umaku.ai project
• Receive notifications and ticket assignments
• Clone and read a linked GitHub repository (read-only)
• Interpret repository structure, documentation, and configuration
• Assemble contextual understanding from project artifacts
• Respond accurately within engineering workflows
• Attempt interpretation of UI elements
• Log ambiguities, failures, and misinterpretations

Timeline

• Sprint 1 (Weeks 1–2): Foundation & Core Integration Setup

Objective: Establish a controlled experimentation environment and validate system connectivity.

• Sprint 2 (Weeks 3–4): Context Assembly & Ticket Interaction

Objective: Enable contextual understanding and workflow participation.

• Sprint 3 (Weeks 5–6): Experimental UI Understanding & Robustness

Objective: Extend interpretation capabilities and improve reasoning stability.

• Sprint 4 (Weeks 7–8): Evaluation & Architectural Conclusions

Objective: Convert experimentation into structured insight.

This project is a controlled research initiative designed to validate what is technically feasible today with autonomous agentic AI inside real-world software engineering environments.

**More details will be shared with the designated team.