Professional software developers don't vibe, they control

Study Methodology and Validity

• Several commenters question the small sample (13 observations, 99 survey responses) as “not statistically significant.”
• Others note this is qualitative research, where concepts like data/thematic saturation matter more than p‑values.
• Some remain skeptical of “qualitative methods papers” as potentially narrative‑driven.
• There’s concern that LLM research ages quickly: fieldwork from mid‑2025 might already lag current models/tools, though others argue findings remain valid if changes are incremental.
• One commenter is uneasy that the lead author has many recent preprints, reading that as possible corner‑cutting.

From “Vibing” to Controlling Agents

• Many resonate with the idea that senior work is less about typing and more about steering systems—LLMs just make this explicit.
• A recurring complaint is mental exhaustion: constant intent‑communication, review, and waiting on agents feels like middle management rather than IC work; flow states are rarer.
• Others see little change relative to tech lead/architect roles, where “empowering others to deliver” was already the main job.
• There’s debate over whether agent supervision plus tests/architecture is more or less efficient than just coding directly, assuming high expertise.

Productivity, Quality, and New Abstractions

• Some report huge speedups (“jetpack for my mind,” multiple prototypes in days), especially for greenfield or hobby‑scale projects.
• Others see agents generating superficially working but structurally poor code: N+1 queries, tangled conditionals, weak permission checks, design rot.
• One pattern: agents excel at boilerplate, glue, and smaller modules; they struggle with very large, long‑lived, domain‑heavy systems.
• Tests—especially integration/E2E—are framed as the new key abstraction and long‑lived context that “fences in” agent behavior.
• There’s disagreement about using agents for planning/design: some enjoy back‑and‑forth architectural work; others distrust AI at that level.

Divided Developer Attitudes

• Clear split between “craft/decomposition lovers” who enjoy understanding every layer vs “black‑box/outcome‑only” users delighted to ship more with agents.
• Some insist they’ll continue hand‑coding for fun even if it hurts employability; others call that naive or “privileged,” emphasizing economic realities.
• Many fear the profession is hastening its own commoditization, while others analogize to open source: huge productivity gains didn’t kill demand, though there’s debate about long‑run wage effects.

Security, Definitions, and Culture

• Concerns about secrets in repos and .env files interacting with AI tools; some argue proper practices and .gitignore mitigate this, others remain wary.
• “Agent” is defined as an IDE/terminal‑integrated tool that can edit code and run tests, vs web chat.
• Some are tired of the “vibe coding” framing and hypey, AI‑styled titles, seeing them as clickbait or reinforcing “real programmer” tropes.