The paper evaluates vibe coding, not agentic systems.

The state of the art is spec-driven development - the agents are given a plan (which could be also iterated and refined with same or different agents).

Useful science related references (among many others) :

• https://arxiv.org/abs/2602.00180
• https://journals.plos.org/ploscompbiol/article?id=10.1371%2Fjournal.pcbi.1014428
• https://www.infoworld.com/article/4166817/vibe-coding-or-spec-driven-development-how-to-choose.html