A few weeks ago at a research leadership dinner, someone said something that's been rattling around my head ever since: “don't outsource things that bring you joy.”
While they weren't talking about research, I kept coming back to this idea during our recent Industry Spotlight webinar with Reggie Murphy, Senior Director and Head of UX Research at Zendesk. Reggie walked through how his team is integrating AI without surrendering to it, a phase he identified as “the messy middle.” That's exactly the right name for where most teams are right now: using AI, but not quite trusting it and not sure where the guardrails should go. Reggie laid out three principles for working through it, and here's what I took away:
Separate slow work from hard work
Reggie's first point is that AI doesn't replace the work that is hard; it replaces the work that is slow.
The slow work (summarizing transcripts, formatting screeners, basic coding) can move to AI, not because it's unimportant, but because it's a tax on the time you should spend on the hard stuff: synthesis, judgment calls, the strategic "so what."
My addition to Reggie's framing is that some of the slow work is work you actually love. The early sifting through transcripts where you catch an unexpected thread; the moment a pattern you weren't looking for starts to surface. AI can do that faster, but faster isn't always better when what you lose in the process is the discovery itself.
I'm not saying don't automate. I'm saying be intentional about what you hand off. Don't outsource things that bring you joy, or at least know you're making that tradeoff consciously.
The false fluency trap
This is where Reggie really got my attention.
False fluency is when AI sounds completely authoritative while being completely wrong. It doesn't hedge. It doesn't say "I'm not sure about this." It produces a finding, in your format, with your confidence level that most of the time nobody checks.
Reggie cited a 2026 MeasuringU study that stopped me cold: of 11 AI-generated UI findings that hadn't been human-verified, 10 were false alarms or outright hallucinations.One was real.
This is something we think about constantly at Sprig as we build out our synthesis tools. Garbage in, garbage out is the old version of this problem. The new version is more insidious. The output looks clean even when the foundation is shaky. The output is not obviously wrong, it's confidently, plausibly, presentably wrong. And if the researcher isn't bringing their own context (the business, the research question, the specific user population, etc.), what comes back is polished, but hollow.
Reggie's name for the antidote is the "trust architect." The researcher's job becomes about systematically asking the questions that AI speed wants to skip. This does not require slowing everything down but knowing exactly which moments require a human to stop and verify before the insight moves upstream.
Architect where judgment lives
You can't fight false fluency by just telling your team to be more careful. What Reggie advocates, and what I'd push further, is mapping your workflow and being explicit about where human checkpoints should live. At what point does an AI output become an organizational decision, and who owns that handoff?
A few things that work: treat the LLM as a sparring partner, not a ghostwriter. Ask it to poke holes in your argument before you present it. Chase the outlier, not the pattern. AI flattens toward the center, and the interesting stuff is usually at the edges. And before anything AI-generated reaches leadership, a human should review it.
What comes next
Reggie's framework is the right map for the messy middle, and I think the next conversation will be about what we're building toward on the other side of it.
The goal will be to get clearer than ever about what human judgment actually is, what it looks like in practice, where it lives in the workflow, and how we develop it in the next generation of researchers who are learning this job with AI already in the room.
That part I'm still working through and that I suspect most of us are too.
Watch the on-demand webinar, The Messy Middle of Craft and Speed, to see Reggie's frameworks and real AI prompting examples in full.
.png)