Automation of the Software Development Process
The research report analyzes the automation of the software development process and the impact of AI coding agents on the efficiency, quality and organization of project work. The case study is based on Project Venom and the GH-2026 and SQ-2026 datasets from Q1 2026.
Key research findings
The publication version guides the reader from the thesis and data to the process model, results and research limitations.
Scale of the analyzed project at the end of Q1 2026.
Commits in the analyzed GH-2026 data period.
Issues and 0 days of technical debt at the end of the cycle.
Test coverage with 5,366 unit tests.
What exactly was tested?
The report addresses whether, and to what extent, the use of AI coding agents affects the quality, time and organization of the software development process compared with traditional methods.
Scale and efficiency
AI coding agents make it possible to run a continuous development process and significantly increase pace within a short project cycle.
Quality and stability
A high pace of code generation does not reduce quality if the process includes quality gates, tests, review and technical debt assessment.
Change in the human role
The human role shifts from writing code toward defining scope, architecture, quality control and acceptance of outcomes.
Data sources and interpretation method
GH-2026
Author’s own data collected from the GitHub API for Q1 2026: commits, active days, lines added, lines deleted and the derived code churn indicator.
SQ-2026
Author’s own data from SonarQube Cloud: issues, technical debt, LOC, test coverage and the number of unit tests at process control points.
What does external research show?
This section presents values from a review of external research on AI-assisted coding. It is a reference point for the report, not an interpretation of Project Venom results and not its control group.
Task acceleration
Code generation
Complex tasks
Unit tests
Review / PR quality
Churn / duplication
AI technical debt
Agent scale
AI adoption
for writing code
Project Venom as a case study of a software development process supported by AI coding agents
Venom was treated as an original, open-source experimental software project and as a case study of a development process in which AI coding agents support implementation, while the human remains responsible for architecture, quality control and acceptance of changes.
Agentic system
The project is used to test concepts of agentic systems: digital roles, contextual memory, workflow mechanisms and interfaces for language models and AI components.
Local-first
The system was run locally on a PC-class machine using an NVIDIA RTX 3060 GPU with 12 GB VRAM and local runtimes such as Ollama, vLLM and ONNX.
Division of responsibility
AI performs selected execution stages, while the human retains responsibility for the goal, scope, architecture, quality control and acceptance decisions.
GH-2026 — repository activity
This view uses only the GH-2026 dataset. It compares the scale and pace of work in repositories, not code quality.
Project Venom against contextual projects — GitHub API
Project Venom is highlighted in a dark color. The GH-2026 data shows work volume: commits, active days, lines added, lines deleted and code churn.
SQ-2026 — code quality
This view uses only the SQ-2026 dataset. The set of projects differs from GH-2026, so the chart presents a separate quality comparison.
Project Venom against contextual projects — SonarQube Cloud
The chart shows the start and end values for contextual projects from SQ-2026. For Venom, the key change is the direction of movement: reduced issues and technical debt, and increased test coverage.
Full software development process of Project Venom
The process model shows how software development process automation was embedded in four responsibility layers. AI accelerates implementation and technical validation, but boundary decisions remain with the human: from goal and scope to architecture, review, merge and assessment of business impact.
Key results of Project Venom
This section brings together the most important GH-2026 quantitative data, SQ-2026 quality data and supplementary observations for Project Venom only.
Project Venom — process scale and quality change
The chart does not compare Venom with other projects. It is a synthetic view of work volume from the GitHub API and the quality trajectory from SonarQube Cloud.
| Area | Initial value | Final value / result | Interpretation |
|---|---|---|---|
| Commits | — | 1,587 | High activity in the short Q1 2026 cycle. |
| Active days | — | 64 | Continuity of the development process. |
| PR lead time | — | average 2.2 h; median 0.9 h; merge rate 95% | Supplementary observation on time economics and a short delivery cycle. |
| Lines added / deleted | — | 875,377 / 332,208 | High volume of changes with parallel quality control. |
| Code churn | — | 38.0% | The process included intensive restructuring and refactoring, not only simple code growth. |
| Issues | 1,650 | 0 | Closure of quality issues in SonarQube Cloud. |
| Technical debt | 19 days | 0 days | Reduction of debt to zero in the adopted metric. |
| Test coverage | 68.7% | 92.2% | Increased regression control and change stability. |
| Unit tests | 1,391 | 5,366 | Expansion of the automated validation layer. |
| Final codebase | 104,950 LOC | 138,011 LOC | Increase in project scale while closing quality gaps. |
| Cost of AI tools | approx. USD 100 / month | approx. USD 200 / month | Economic observation; the study does not include a full TCO calculation. |
What follows from the study?
H1 confirmed
The scale of 138,011 LOC, 1,587 commits, 64 active days and 38.0% code churn shows an intensive, continuous process over a short period.
H2 confirmed
The high pace did not lead to a lasting reduction in technical quality because the process included measurement, quality gates, tests, review and debt control.
H3 confirmed
The human moved toward a decision-making role: goal, scope, architecture, acceptance and responsibility for the result.
Boundaries of interpretation
Single operator
The work is an analysis of a single case carried out by one operator. The conclusions should not be automatically transferred to multi-person teams, other domains or corporate environments.
Nature of the metrics
LOC, technical debt and test coverage are quantitative technical metrics. They are not direct measures of business value or long-term architectural stability.
Economic scope
The economic analysis covers direct subscription costs of AI tools, approximately USD 100 → 200 per month. It is not a full TCO calculation.
Frequently asked questions about the study
Research FAQ
This short section organizes the key questions about software development process automation, Project Venom and the role of AI coding agents.