Our Qualicon 2026 panel on AI in QA generated more audience questions than we had time to answer live, so we brought Rex Black, Andrew Fray, Ionut Codreanu, and Aniruddha Pawar back for a follow-up to work through the rest of the audience questions from Qualicon 2026. Below are some key points and timestamps from this panel.
If you haven’t already, you can stream the first part from this year’s Qualicon in our members only vault.
AI helps you think, it doesn’t decide for you.
Aniruddha described using AI to get unstuck on structured problem-solving exercises like abstraction laddering, but the output always needed human validation before it could be trusted.
Hallucinations don’t go away just because you ask nicely.
Andrew shared a case where an AI tool invented a version number outright, admitted to fabricating it when caught, then later hallucinated that a “no fabrication” rule had been active far longer than it actually had.
Data governance and optics matter.
Ionut and Aniruddha flagged data storage and long-term tool ownership as real risks, while Andrew recounted a studio’s rough week after players found a leftover AI config file in a shipped build.
The real upside is new capability, not just speed.
Andrew argued coding agents matter most for letting designers and artists build their own tools, freeing them from waiting on engineers.
Real use cases that are already paying off: build failure triage, changelist-based test targeting, and small tools quietly killing repetitive busywork.
Key Timestamps:
0:29 — Intro to the bonus Q&A and first question on building process models with AI
0:57–3:55 — AI for structured problem-solving: helpful for unblocking, but outputs need heavy human verification
4:14–7:06 — Risks of using AI in QA/game dev: maintenance, governance, data security, and provider dependency
7:45–9:39 — Mitigating provider risk with backups/local LLMs and protecting confidential data
10:34–12:03 — Public reaction to AI use and disclosure concerns
13:44–17:57 — How to avoid bad practices caused by low-cost AI actions; metrics and incentive design
18:15–19:40 — Hallucination/fabrication example and the need for explicit no-hallucination rules
23:57–30:03 — How to get started with AI: start small, use low-risk repetitive tasks, build skills gradually
31:16–39:43 — AI blurring role boundaries: opportunity vs threat to specialized professions
41:49–44:30 — Security best practices: safe environments, layered validation, strict access control
45:29–48:46 — Avoid overcomplicating AI output and don’t mistake token usage for productivity
49:01–51:46 — Lightning round: top AI use cases from each speaker
For more updates, head to our news section!