Reasoning effort, not tool access, buys first-try reliability in agentic code generation: an observational study

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Achint Mehta
• Venue: arXiv: cs.SE
• Published: 2026-07-02
• Collected: 2026-07-03
• URL: http://arxiv.org/abs/2607.02436v1
• PDF: https://arxiv.org/pdf/2607.02436v1
• DOI: UNKNOWN

Plain-Language Brief

A study compared different agentic coding assistants and their abilities in building a real-time retrospective board. The results showed that the level of reasoning effort, rather than access to additional tools or capabilities, significantly influenced the reliability of the first try. A higher reasoning effort led to fewer errors and less need for corrective prompts.

What the Document Says

• Agentic coding assistants were tested in building a real-time retrospective board
• 90 independent runs were made, each scored on a 14 criterion functional rubric
• The runs spanned several model generations, two agent harnesses, two reasoning effort levels, a testing tool, and two design oriented prompts
• Frontier models performed better than a low cost local model
• Container deployment was the dominant defect, failing in 44% of runs across model generations
• The testing tool increased cost by 42% to 68%, but did not improve functional score or reliability
• Raising reasoning effort improved first try perfect runs and reduced corrective prompts
• A design oriented prompt improved visual quality without lifting function
• A one paragraph paraphrase of the directive's directive reproduced the entire lift

Key Concepts Mentioned

• Agentic coding assistants
• Real-time retrospective board
• Model generations
• Reasoning effort levels
• Testing tool
• Design oriented prompts
• Container deployment
• Functional rubric

Claims Made by the Document

• More capability does not necessarily yield better software
• Higher reasoning effort leads to fewer errors and less need for corrective prompts
• Container deployment is a dominant defect in the tested runs
• The testing tool does not improve functional score or reliability
• A stronger model or more effort can prevent first run failures due to weak reasoning

Evidence Used

• Comparison of scores and failure rates among different models and tools
• Analysis of defects in container deployment across model generations
• Results showing the impact of reasoning effort on errors and corrective prompts
• Data indicating the lack of improvement in functional score or reliability with the testing tool

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

Agentic coding assistants are increasingly given extra capabilities, such as browser based testing tools and design oriented system prompts, on the assumption that more capability yields better software. This study tested that assumption directly. Ninety independent agent runs built the same application, a real time retrospective board, from one detailed specification, each scored on a fixed 14 criterion functional rubric (42 point maximum) and a visual quality review. The runs spanned several model generations, two agent harnesses, two reasoning effort levels, a testing tool, and two design oriented prompts. Capability tier dominated: frontier models clustered near the ceiling while a low cost local model fell to 24 to 37 points. A criterion level analysis revealed what run totals conceal. Container deployment was the dominant defect, failing first try in 44 percent of runs, with its failure rate shifting sharply across model generations while mean totals moved less than a point. The testing tool raised cost by 42 to 68 percent without improving functional score or reliability, even on interface visible criteria. Raising reasoning effort from High to xHigh lifted first try perfect runs from 28 percent to 89 percent and cut corrective prompts about five fold, for 9 to 29 percent more cost. A design oriented prompt raised visual quality, 4.5 versus 3.0 on a 5 point scale, without lifting function, and a one paragraph paraphrase of its directive reproduced the entire lift. The practical lesson is to match the fix to the failure: most first run failures came from weak reasoning, which a stronger model or more effort prevents, not from visible flaws a checking tool would catch.