The death of the writer, the birth of the orchestrator?
Dr Susanne Friese · · Methodological debates
If you read my last newsletters, you know the AIAgents4Qual conference took place recently. Let me share a bit of what we learned.
We asked a radical question: What happens if we let AI take the lead? The answer was immediate and sobering: it fails.
While AI can mirror the sound of a scholar, it cannot sustain the logic of one. Our contributors found that "persuasive form" arrived far faster than "warranted substance". The AI produced elegant prose that masked a hollow evidentiary core, fabricated citations, and missed the "close, resistant reading" that defines qualitative inquiry.
The lesson is not that we should become better prompt-engineers, but that we must remain the primary "meaning-makers". AI is a powerful "irritant" and an accelerator for low-level tasks, but it lacks the lived experience, ethical weight, and reflexive depth required to anchor a study.
Below I put together a list of Best Practices for AI-Augmented Authorship
Beyond the Prompt: A Guide to Responsible AI Collaboration Based on the autoethnographic reflections of researchers at the 2026 AIAgents4Qual conference, here are the emerging "gold standards" for maintaining integrity in AI-supported work: • Adopt the "Orchestrator" Mindset: Move away from delegating the "analytic agency" entirely. Instead, treat the human role as one of "methodological governance"—constantly auditing and disciplining the AI’s output. • Beware of "Persuasive Form": Recognize that AI is optimized for "surface fluency". High rhetorical polish often masks "weakly supported claims" or "unresolved assumptions". • Mandate Evidence Anchoring: Force "claim-support tables" or "audit trails" early in the workflow. Never allow synthesis prose to be drafted before the evidence is explicitly anchored to the raw data. • Engineer Epistemic Friction: Do not ask for a single "winner" or a "clean" conclusion. Use "persona differentiation" to assign roles like "adversarial reviewer" or "diagnostician" to force the AI to surface contradictions and "view-from-nowhere risks". • Verify at the Source: Treat AI-generated citations and page numbers as "hallucination risks" by default. Implement a strict rule that no content-specific interpretation is retained unless confirmed directly in the original document. • Disclose the "Context Engineering": Transparency is no longer just about the final data; it is about the "generative conditions". Document the prompts, the model priors, and the "role architecture" as part of the formal research record
The presentation recordings can still be viewed. The conference platform stays open until April 13, 2026. Same agenda, same sessions. Pick what interests you and watch on demand, at your own pace: