CARBON · AI testing inside your coding agent.
Give your AI testing a framework: explore risks, run checks, and review the evidence.
Explore CARBON ↗AI testing / Research results
What works.
What’s missing.
What comes next.
A shared view of how people use AI for testing, where expert judgment matters, and what they want to change.
Written feedback
Keep the context. Protect the people.
Free-text answers are not reproduced in this small-sample public snapshot. No names, emails, individual response records or follow-up preferences are embedded.
How to read these results.
Each chart shows percentages of people who answered that question. Unanswered questions are excluded from percentages; “not assessed” and “not sure” remain visible answers. Multi-select percentages can exceed 100%. Only selected options are shown; unlisted options received no selections.
This is a small, self-selected sample of reported experience—not measured defect detection, an AI miss rate, or evidence of causation. These responses do not establish causation or population trends. Book and summary requests require complete applicable answers; other submissions may skip questions.
Spreadsheet snapshot · Not auto-refreshing
Aggregated from the private response Sheet on October 15, 2026. Repeat email submissions are consolidated using the latest response; responses without email remain separate. The Sheet remains private. Refresh this file when more responses arrive.
Aggregated from the private response Sheet on October 15, 2026. Repeat email submissions are consolidated using the latest response; responses without email remain separate. The Sheet remains private. Refresh this file when more responses arrive.
How much might the percentages move?
Compared with the responses available before October 8, the median absolute change across currently selected chart options is 1.1 percentage points; the largest is 7.2 points. These are overlapping snapshots, not independent samples or proof of future stability.
Confidence intervals and their limits
For questions with the current response bases, a hypothetical simple random sample would have a worst-case 95% interval of roughly ±9.0–9.4 percentage points. This is an illustration, not a valid population margin of error for this opt-in survey. Selection bias is not included, and intervals do not predict how future answers will move.
Illustration uses 1.96 × √(0.5 × 0.5 / question response base). No weighting is applied. It assumes independent, representative random draws—assumptions this opt-in survey does not establish. Survey-method guidance from AAPOR.
Bookmark this results URL. We will refresh this snapshot as more responses arrive; it does not update automatically.
AI tools. Experienced QA experts.
Add AI-powered testing bandwidth, review the findings, and help your team learn to run it themselves.
Talk to IcebergQA ↗More testing throughput. Let’s work together.
Want help or collaboration? Whether you need AI testing tools, AI QA experts, or both, we can help you increase testing throughput and make better release decisions.
Let us know → jason.arbon@icebergqa.com