TL;DR
A December 2025 study analyzed how experienced software developers integrate AI agents into coding workflows. Using field observations and a qualitative survey, the authors find developers treat agents as productivity tools but preserve human control to protect software quality.
What happened
A team of researchers led by Ruanqianqian Huang and colleagues examined how professional developers use AI agents when building software. The study combined field observations of 13 developers with a qualitative survey of 99 practitioners to explore motivations, strategies, task suitability, and sentiment toward agent use. Findings indicate that experienced developers appreciate agents for boosting productivity but do not cede core design and implementation decisions. Instead, participants insisted on maintaining control to ensure fundamental software quality attributes and reported deliberately applying their expertise to steer and constrain agent behavior. Overall sentiment toward agent integration was positive, with developers expressing confidence that human oversight complements current agent limitations. The authors conclude that established software engineering practices remain important when working with agents and highlight opportunities for improved agentic interfaces and guidelines for practitioner use.
Why it matters
- AI agents can speed routine coding tasks but do not replace developer judgment on design and quality.
- Maintaining human control helps preserve software quality attributes that practitioners prioritize.
- Findings underline the continued relevance of software engineering best practices in agent-assisted workflows.
- The study points to product and research opportunities around agent interfaces and usage guidelines for professionals.
Key facts
- Paper title: "Professional Software Developers Don't Vibe, They Control: AI Agent Use for Coding in 2025."
- Authors include Ruanqianqian Huang, Avery Reyna, Sorin Lerner, Haijun Xia, and Brian Hempel.
- Methods: field observations (N=13) and a qualitative survey (N=99).
- Experienced developers reported AI agents provide a productivity boost.
- Developers retain decisional control over software design and implementation to protect quality.
- Participants use expertise-driven strategies to control agent behavior (specific strategies not detailed in the abstract).
- Overall developer sentiment about incorporating agents was positive, based on confidence in complementing agent limitations.
- The authors suggest future directions: better agentic interfaces and agentic use guidelines for practitioners.
- Paper posted to arXiv on 16 December 2025 (arXiv:2512.14012) with a DOI link provided in the record.
What to watch next
- Development of specialized agentic interfaces tailored to professional software workflows (noted as a future opportunity in the study).
- Creation and adoption of agentic use guidelines for developers (identified by the authors as a research and design direction).
- Impact of agents on large-team design processes and long-term maintenance costs: not confirmed in the source.
- Specific tactics developers use to control agent outputs in practice: not confirmed in the source.
Quick glossary
- AI agent: A software system that performs tasks or makes suggestions autonomously or semi-autonomously, often using machine learning or large language models.
- Field observation: A qualitative research method where researchers watch and record participants' behavior in real-world settings.
- Qualitative survey: A survey that collects non-numeric data such as opinions, descriptions, or experiences to understand participant perspectives.
- Software quality attributes: Characteristics like correctness, maintainability, performance, and security that define the overall quality of a software system.
- Agentic interface: A user interface designed to let users interact with, direct, and constrain autonomous agents effectively.
Reader FAQ
What did the study investigate?
How experienced developers use AI agents in software development, including motivations, strategies, task suitability, and sentiments.
How many participants were involved?
The study included field observations of 13 developers and a qualitative survey with 99 respondents.
Do developers let agents write entire applications autonomously?
Not confirmed in the source.
Are developers generally positive about using agents?
Yes. The study reports an overall positive sentiment, with developers confident that human oversight complements agent limitations.
Computer Science > Software Engineering [Submitted on 16 Dec 2025] Professional Software Developers Don't Vibe, They Control: AI Agent Use for Coding in 2025 Ruanqianqian Huang, Avery Reyna, Sorin Lerner,…
Sources
- Professional software developers don't vibe, they control
- Professional Software Developers Don't Vibe, They Control
- Prof. Software Developers Don't Vibe, They Control
Related posts
- Zpdf: Zig PDF text extraction library — up to 5x faster than MuPDF
- Everything as Code: How One Company Runs Its Entire Platform in a Monorepo
- Professional Developers Direct, Not ‘Vibe’: AI Agent Coding in 2025