AI now writes a competent research report from a folder of transcripts in minutes. If your research practice was built around producing that report, the uncomfortable news is that the report was never the valuable part.
Synthesis used to be the researcher's visible labour: the affinity map, the themes, the deck. Models do a serviceable version of all of it, faster, and well enough that stakeholders will stop waiting for the human version. Mourning that wastes a good disruption. The better question is what the discipline owns once the summary is free.
i.What dies and what survives
What dies is research as a deliverable: the project-shaped engagement that ends with a readout and a folder nobody reopens. What survives, and grows, is the work the model can't touch. A model can't decide which question is worth asking, because that takes knowing what the organisation is afraid to learn. It can't build access to real users, in real contexts, over time. It can't tell you when the data is lying to you, or when a tidy theme is really an artefact of how you prompted it. And it can't carry an unwelcome finding through an organisation that would much rather not hear it.
So research stops being a deliverable and becomes the evidence loop that keeps judgement honest.
ii.The evidence loop
The practice I'm building treats research as infrastructure rather than as a series of projects: a standing panel of real users, instrumented surfaces, and a regular rhythm of contact, so that any significant product judgement can be checked against evidence at the moment it's made, rather than six weeks later in a readout nobody reopens. AI accelerates every step of that loop. It transcribes, clusters, drafts, cross-references. What it doesn't do is own the loop's integrity, and the integrity is the actual product.
This is also where inferred data earns its scrutiny. When a system learns about users by inference, someone has to ask whether the inference is fair, whether the user would recognise themselves in it, and whether acting on it is acceptable. That question is epistemic and ethical at the same time, and it lands on research, because nobody else in the building is paid to care whether we actually know what we claim to know about the people we build for.
iii.The researcher as epistemics owner
The title I'd give the discipline now isn't synthesiser. It's epistemics owner: the person accountable for the gap between what the organisation believes and what's actually true of its users. The report was only ever a vehicle for that accountability. The vehicle has been automated. The accountability hasn't been, and in an era where every team can generate confident output on demand, it's worth more than it ever was. ✳