AutoCodeRover: from research on Automatic Programming to Spinoff Acquisition
AutoCodeRover: from research on Automatic Programming to Spinoff Acquisition
Agent for Autonomous Program Improvement
Featured: AutoCodeRover
Agent for Autonomous Program Improvement
AutoCodeRover: from Research on Automatic Programming to Spinoff Acquisition
AutoCodeRover: from Research on Automatic Programming to Spinoff Acquisition
Agentic AI Software Engineer: Programming with Trust
Agentic AI Software Engineer: Programming with Trust
Professor Abhik Roychoudhury Comments in IEEE Spectrum
Professor Abhik Roychoudhury Comments in IEEE Spectrum on Importance of Intent Inference and Trust in AI Coding

Agentic AI based Software of the future: from Scale to Trust

NRF Investigator funded by the National Research Foundation, Singapore (2026-31)

Agent Future Workflow

Figure: An envisioned workflow for future software development with agentic AI.

Overview

The practice of software is significantly shifting due to the rapid development and improvement of generative artificial intelligence (GenAI). In particular, large language models (LLMs) have shown significant promise, especially when deployed as LLM agents, i.e., when LLMs are combined with external tools to aid decision-making and action-taking. The unprecedented pace of change makes it ever more important to understand how software engineering will change, and what impact this will have on industry.

This research program aims to examine how future software engineering is likely to shift from scale to trust as agents become prevalent. Prior software engineering research emphasized how a large number of generally trustworthy human developers could coordinate to make large software (scale). However, we foresee a change in which often untrustworthy but fast agents will write a large proportion of future software, making it important to research how to increase trust. More specifically, we posit that it is important to engender trust in software artifacts such as code automatically generated by AI agents.

Goal

In pursuit of the vision of shifting research from scale to trust, we will research trust-building technology that would automatically (1) improve the quality of agent-generated software, and (2) generate additional artifacts to enhance the trust developers can place in agent-generated software. We propose to research a unification of specialized agents to a single unified software agent that can handle all tasks, and a mechanism for validating AI-generated software comprehensively so that AI artifacts come with degrees of assurance. Overall, an AI workflow for software systems of the future would be delivered; this workflow would be deployed at enterprises to not only speed development, but make it trustworthy.

Initial Work

Unified Software Engineering agent as AI Software Engineer, Leonhard Applis, Yuntong Zhang, Shanchao Liang, Nan Jiang, Lin Tan, Abhik Roychoudhury, ACM/IEEE International Conference on Software Engineering, 2026.

Open Positions

We are immediately hiring 1 post-doctoral researcher based at NUS working on building trusted agents. Interested applicants can send their CV to vivien@nus.edu.sg with CC to Prof. Abhik Roychoudhury. Details will be discussed with interested applicants only.

News