Before AI ever showed up, Tina’s route into insights was already a clue about where the real value sits. “I’m a marketer at heart,” she says, and the pivot happened when a research leader framed the job as “strategic thought partners than just research delivery.”  

That is prompt engineering in human form. The part where someone refuses to accept a vague brief, pushes for what the business is really trying to do, and turns that into questions that can survive contact with reality. 

It is also why the “prompt” can’t just be a clever line typed into a chat box. In a real organization, the prompt is the whole setup around the question. Definitions. Comparability. The boring stuff. The stuff that stops your tracker from becoming an argument generator. 

The episode drifts into this territory when the talk turns to data strategy. One line sticks because it is exactly what goes wrong in big companies. “One source, one truth,” and then the important nuance, “we’re not just going to change the definition to make it look better in a graph.”  

That is the hidden craft. If your metric changes meaning every time a team touches it, AI won’t save you. It will just accelerate the production of confident-looking nonsense. Prompt engineering, the real kind, starts earlier. It starts with shared language and standards that hold when the room gets political. 

There’s another shift running underneath the AI conversation that matters more than most tool debates. 

Brands are no longer willing to accept mystery numbers. JT describes the “balance of power moving from the agency to the brand,” with black-box IP “being undone rapidly as brands want to know why that number moves?”  

That demand changes everything. Once stakeholders ask, “What do I have to do to understand it?” you can’t hide behind a dashboard. You need traceability. You need repeatability. You need to explain movement without sounding like you’re translating scripture. 

That, again, is prompt engineering. It is the discipline of making your work legible. And it shows up in small choices that feel unglamorous until they save you. Defining what a segment means and keeping it stable. Knowing what “channel” means in your data and not quietly rewriting it when performance dips. Building the foundations so the question doesn’t change every time the answer becomes inconvenient.  

One of the most practical parts of the conversation is the way Tina talks about maturity. 

Some companies are so resourced that the supplier side can feel small by comparison. JT gives the example that one client “has doubled the number of data scientists in their organization than we have in our entire staff.”  

There are also companies with “two researchers in the whole company,” and what they need is completely different, adds Tina. 

That matters because it kills the fantasy that AI will level the playing field by default. 

Tools don’t remove complexity. They move it around. In mature organizations, AI tends to amplify speed and scale, but also raises the bar on governance and explainability. In lean organizations, AI can feel like an extra set of hands, but only if someone knows how to steer it and spot when it is drifting. 

Either way, the advantage is not “who adopted the newest tool.” The advantage is who can frame the problem cleanly, hold definitions steady, and translate outputs into something the business can actually use. In other words, who can prompt. 

Machines can get surprisingly far. Then they fall down. “It does hallucinate. So it does fall down.”  

So the answer isn’t that everyone becomes technical, or that nobody needs expertise. The answer is the same pattern research has always relied on. Deep experts “available when things don’t look right,” and generalists who can “turn the handle of the machine.”  

That matters because it reframes what “prompt engineering” actually means inside a team. 

It’s not a lone power user with a secret prompt library. It’s a function that can run fast without losing its standards. A system where someone can move work through efficiently, and someone else can step in when the output looks too neat, too certain, too clean to be trusted. 

“We’re just rebuilding that machine at the moment.”  

If you want the simplest description of what will separate strong insight teams from the ones that get swallowed by tool churn, it shows up right at the end. 

“The number one skill… is a growth mindset.”  

Not because it sounds good on a slide. Because the teams that cling to “I’m an expert in blank” stop learning right when learning becomes the job.  

Then comes the core of prompt engineering, without the buzzword. The ability to force the uncomfortable question. “What is it that we’re trying to accomplish here,” with the patience to go deep, and “the courage to force that conversation because it’s tough sometimes.”  

And then the part nobody gets to skip, no matter how good the tools get. “Interpretation for the business,” the storytelling, the influence, the relationships that make the work travel.  

That is why prompt engineering was always the job. Not because researchers were secretly doing computer science. Because the job has always been about directing attention, protecting meaning, and turning messy reality into a question sequence that leads to a decision. 

And the best closing line in the whole episode is also the most freeing. “No one out there is an expert… none of us know what we’re doing.”  

Good. That means you don’t need to wait to become one. You just need to do what insights has always done at its best. Stay curious. Keep your standards. Ask better questions than everyone else in the room. Then keep tightening them until the answers are something a business can actually act on. 

You can watch or listen to the full podcast episode here: https://www.research-revolutionaries.com/e13-cheaper-faster-but-is-it-better-ais-impact-on-consumer-research/