Every networking break at our Experience Research Summit in Silicon Valley ran long. We'd call the room back for our next speaker session, and the conversations wouldn't come to an end. Research leaders from Google, Microsoft, Figma, Salesforce, Uber, Workday and Atlassian compared with their peers how they're actually using AI with their teams, which is not an opportunity you get at your average conference talk or in a comment thread. These 50 research leaders all had the same challenge with no space to talk candidly about it, and the second they were given one, they didn't want to leave.
Here's the topics these researchers were discussing that kept networking breaks running long.
The unfair paradox
Research teams are receiving more questions than any team can answer at depth, and there are leaders making expensive calls on instinct because waiting for evidence would cost them the quarter.
While the natural instinct is to solve that problem with headcount, it doesn't work. The product org will always out-hire you, out-pace you, and out-ask you. Real scale is decentralized: the tactical, evaluative questions go to the teams asking them (oes this layout make sense? A or B? Is this copy clear?), and you keep the ambiguous, generative work that shapes the roadmap. This works just fine, right up until someone does it without guardrails.
Bad data gains unearned credibility the moment the word "research" appears on the cover slide. Open the faucet before you've laid the pipe and you don't get democratization; you get a well-formatted mess with your team's name on it.
Lay the pipe first, then the faucet. Create vetted assets, standardized methods, a searchable place people check before they study, and expert sign-off before anything goes live.
Three kinds of research leader
Many research leaders already have these checks in place but are still struggling with the pace of demand because their workflows are so manual. In the current AI era, I have seen three types of research leaders emerge:
- The Resister blocks AI tools in protection of the craft, and then watches the organization quietly route around them.
- The Delegator hands everything to AI, keeps no oversight, and inherits beautifully formatted mistakes.
- The Orchestrator designs the system and governs the machine. This is the move from being in the loop to being on the loop.
Most of the room put themselves somewhere between the first two types of leaders and wanted to know how to get to the third. Fair.
Why the third one is hard to get to
An AI agent can architect a study and inspect it for broken logic, dead ends, and leading questions, another can run collection and adapt as answers come in, and a final can surface themes and draft reports with verbatims still traceable. At Sprig we call those Design, Field and Synthesize agents, and the pattern is becoming standard regardless of whose tools you use. Weeks compress into hours on the same headcount.
What’s vital to becoming an Orchestrator is what comes after those mechanics: your judgment.
High polish can dangerously register as high truth; a clean report with tidy charts reads as more credible than a messy one, whether or not it deserves to. A generic model has no idea that one loud outlier isn't fifteen users, and once a well-formatted insight gets democratized, it's very difficult to walk back. Orchestration is a discipline, not a software purchase, that involves retraining, a review load, and perhaps some uncomfortable months where you're slower before you're faster. Most Orchestrators got there by way of the Delegator, handing too much over first and building the checks afterward. The shift isn't about better tooling, it's about getting explicit on which decisions still need a human before an insight travels any further.
What the room added
I hosted more than I talked, which was the point. After the keynote I sat down with Sanya Attari for a fireside chat, and Mani Pande shared a presentation before joining me for another discussion.
Sanya made the case for a role most research teams didn't have 18 months ago: the AI Ambassador. Someone who sits between research and research ops, spots which recurring questions genuinely deserve an agent, gets it built, then owns whether it keeps earning its place. This role is not a gatekeeper or a manager of the research backlog. When you hire for it, she argued, you're looking for curiosity and orchestration over tool fluency because the hard skill is deciding what to hand off and catching the output when it's wrong.
Mani argued the double diamond is stretching. Researchers have camped in the first diamond, discovery, for a long time, and prototyping got cheap enough that staying there is now a choice rather than a constraint. She argued research should start with more research, mining what the business already knows before anyone writes a plan, so a team spends its time on the genuinely unknown. And the piece worth protecting is storytelling, because models flatten toward the average and have no idea why a study matters to a roadmap.
Then, we put the slides away and sat at five tables, ten leaders each, for sixty minutes with no presenting. The thing I cared about most is that nobody left that hour without having said one concrete thing out loud, because we close on the same question every time: what's one conversation about your team and AI that you'll have differently in the next two weeks?
Join us for Experience Research Summit: Seattle. Wednesday, September 30 at The Edgewater Hotel. Fifty seats are open for VP, Director and Head of Research and Insights leaders to apply. We cap our events deliberately so every conversation in the room is between peers, and so the breaks can run long. Join us for a keynote, two fireside chats, and an hour at a table of ten where you leave with something you've committed to out loud.
Apply for an invitation →
.png)
.png)