How AI is transforming daily workflows: why exposure isn’t adoption and how to make AI stick

Most companies do not have an AI access problem. They have an adoption problem. In this episode of Intelligence: Real & Imagined, Rebecca Hinds, Head of the Work AI Institute at Glean, talks with Shweta Puri, Marketing, Technology and AI Operations Lead at Nextdoor, and Nichole Sterling, Co-founder and Board Member of Women Defining AI and CEO of My Town AI, about what actually moves AI adoption from hype to habit.

What you’ll learn:
• How Nextdoor moved from a 50-person pilot to 500+ users, 700 agents, and strong stickiness by embedding AI in the flow of work
• Why AI champions, peer credibility, and OKRs matter more than company-wide announcements
• How to spot AI theater and replace it with experimentation, better workflows, and measurable habits
• Practical examples of evergreen and seasonal AI agents, from IT help to performance review support

Resources:
• AI Transformation 100: https://glean-it.com/4cMaOgK 
• See the latest from the Work AI Institute at Glean: https://glean-it.com/3NQD33S

Subscribe for more:
https://www.youtube.com/@gleanwork

Follow us:
• LinkedIn: https://www.linkedin.com/company/gleanwork/
• Twitter/X: https://x.com/glean
• Instagram: https://www.instagram.com/gleanwork/

Timestamps:
00:00 – Why AI adoption is harder than AI access
04:15 – From pilot to 500+ users, 700 agents, and 82% stickiness
07:42 – The ambition-execution gap, champions, and OKRs
12:36 – What makes an effective AI champion
18:54 – Measuring success and spotting AI theater
23:13 – Workflow examples: IT help, benefits, and performance review agents

Transcript

SPEAKER_02 0:00

Welcome back to Intelligence, Real and Imagined from the Work AI Institute at Glean. This is the show where we sort through what's real, what's hype, and what actually works with AI at work. I'm your host, Rebecca Heinz, and I lead our Work AI Institute here at Glean. Today I'm joined by two incredible female AI leaders, Shueta Purry from Nextdoor, and Nicole Sterling from Women Defining AI. This episode is inspired by our AI Transformation 100 report, and we're talking about AI adoption, how it happens, not through hype, but through trust, experimentation, better workflows, and people who are willing to rethink how work gets done. Let's dive in. Shweta, why don't you kick us off and please introduce yourself?

SPEAKER_00 0:54

Hi everyone. I'm Shweta Puri, Marketing Technology and AI operations lead at Nextdoor, the neighborhood network connecting people to local communities across the US and beyond. My background is at the intersection of marketing technology operations and more recently enterprise AI. What that practically means is I'm not a researcher or a vendor. I'm a practitioner who has actually had to figure out how to make AI real inside a company with real budget constraints, real skeptics, and real workflows that don't pause for a pilot. So I've been a clean customer and also a power user since our very early days with the platform. And it's been a year or so. I have somewhat unusual kind of vantage point here. I've lifted both sides, the person rolling it out and also the person using it every day. So the dual perspective is what I'm hoping to bring to this conversation today. And I've also written this journey in a Glean's in Glean's blog, which I'll share as a follow-up resource. But the short version is that the technology was the easy part. Everything else is what we are here to talk about. I'll pass it back to Rebecca.

SPEAKER_02 2:07

Wonderful. And I'm I'm so excited for that blog. I'm sure we'll send it as a follow-up. It's it's very insightful. Uh Nicole, Nicole, please introduce yourself.

SPEAKER_01 2:16

Hello, everyone. My name is Nicole Sterling. I am the co-founder of Women Defining AI. We are an AI and education company for women and non-binary individuals. We started several years ago uh as just a lunch and learn, and now we have a community that's nearly 2,000 members strong. And uh I've stepped away from the operational role within women defining AI because I have started my own AI company. This is one of the beautiful things that we've seen come out of our women defining AI community, is women and uh are getting new jobs, starting new companies. Uh, I happen to be a publicly elected official. I sit as mayor pro tem for my community. And so my AI company has merged my subject matter expertise in local government with my technologist background. And so I come at it from not only uh someone who has had to drive AI amongst a community, but also uh uses AI herself and also from a vendor perspective in how I interact with local governments. Uh I also sit on the GovAI coalition, though, so I help uh local governments procure safe, ethical, and responsible AI all across the United States. So excited for the conversation today and thank you for having me.

SPEAKER_02 3:36

So today's session is all about how AI adoption really happens. And what we consistently see is most organizations don't struggle with access to AI. We often think that AI is fundamentally an access problem. If we give employees this technology, they'll they'll use it. The reality is not as such. And what we see is that conversion from real exposure to AI to real change, exposure to change, is very difficult. How do you get this technology in the hands of people in a way that they're actively using it and using it to their benefit day to day? So, Shweta, one of the things we're seeing, I know you're seeing as well, is more and more organizations are establishing new roles that contain AI in the title. And your role is an example of that. Um, marketing technology and AI operations lead. Tell us a little bit about your role, what you're responsible for, and the significance of having AI in the title.

SPEAKER_00 4:41

Yeah, I know. Uh it's exciting times. Um, everybody wants to uh embrace AI in some way or the other, but I'll give you uh uh the perspective from my uh from my side. I've spent the 10 plus uh years kind of leading Martec in operations, which is again not so glamorous uh connective tissue work of making tools actually work for people. So last year I pivoted into AI operations and enablement at Nextdoor, and um Glean was one of our very first platforms that we implemented. What I've come to appreciate about having AI in my title is that it's less about deep technical expertise and more about uh like being accountable for real outcomes, helping teams find meaningful ways to work differently with AI so it becomes a genuine part of how people do their jobs rather than just another tool that we invested in. So my role kind of sits squarely between um business technology and people processes. And the goal is not to deploy more tools, it's to change the behavior or the foundation itself. Um that looked like a practice. Uh, you know, um, we kind of started with 50 people pilot. We didn't just open a sign-up link and hope for the best. We selected the champions who deeply understood the problem. And also that meant from day one, everyone in the room had the context, conviction, and reason to care. So that deliberate curation made all the difference. And within a month, we had a strong uh monthly active user base and also uh quite a few daily active users. So we made the decision to roll it out broadly across the company to around 550 employees, and we are now at 500 plus users, 700 agents built, and almost at 82% stickiness in just six months of that broad rollout. So, yeah, I mean, phase one was getting people to use search, was the entry point and not the destination, and finding answers was useful, but not transformational. So phase two is where things actually started to get interesting when we move for from, you know, I I search from Glean through to like I work through Glean, which is all about agents, automations, AI embedded in the actual job and not like as a side note. Um so yeah, that's that's a you know high-level journey uh so far at Nextdoor.

SPEAKER_02 7:18

And that embedding AI in the flow of work is something we consistently hear is incredibly important. You know, it's sometimes referred to as the toggle tax. And this is something we unpack in our report as well. You know, the reality is employees are often so overwhelmed, so overburdened that they're not wanting to switch to another platform or tool to use the technology. So, Shweta, one of the things I want to double-click on in your journey is what we often see in organizations is what's sometimes referred to as the ambition execution gap. A lot of ambition associated with the technology, a lot of hype. But then when the rubber hits the road, not seeing that translate into real results. Are there aspects of your journey and nextdoor's journey in which you've seen, you know, you've run the training, you've run the pilots, broad rollouts, but there's been a gap in translating that, you know, ambition and hard work and preparation into real results?

SPEAKER_00 8:23

Yeah, that's that's true. I mean, that's a real challenge. Um, and again, training alone um almost never moves the needle. So I would say that as someone uh who has run a lot of it, and here's what I learned the hard way. Think about the last time you actually, you know, changed how you work uh because of a company-wide announcement. Probably never. Like there are always announcements happening in the company, right? But if a peer you respect, someone in a role like you, uh pulled you aside and said, you know, this saved me two hours last week, you would probably try it pretty quickly and drop everything else, right? So AI adoption works exactly the same way. The signal has to come from someone with credibility in the room and not from a central team just kind of broadcasting to everyone. So the second unlock was the OKRs. Uh, when we embedded AI usage into how teams measure success, not as a side uh initiative, but as part of how they hit their goals. AI stopped being, you know, uh an overhead and started being how people actually worked. If your AI program is not touching the OKRs, it's still an optional thing. And people would consider that that as optional, even in their role. So I would say start from the basic, like embed AI into the OKRs or have people kind of um drive more adoption that way.

SPEAKER_02 9:52

Yeah, that's that's great insight for sure. Nicole, I want to turn to you. Have you seen this gap between ambition and execution as well in in your work? And what do you think is is driving it?

SPEAKER_01 10:06

Yes, absolutely. Uh there is always this disconnect between uh leaders who say, we gotta get have that, we have to have AI. And yet they may not, and if they do provide training, that's great. Uh, we've seen a lot of companies across whether it's women defining AI or even um in my uh realm of local government, where they don't provide training, they don't provide guidance. And so the everyone's left wondering, well, well, how how do we do this thing? And I think one of the reasons that's driving it is because there is a lot of hype that, you know, 2025 was all, you know, like a lot about agents. And so everyone is asking, agents, is this something I should know? And and unfortunately, because there is like this pressure towards hype, folks internally they're not doing some of the basic uh, you know, just blocking and tackling of training, experimentation, usage. And so folks don't have a very good understanding of what they could get out of AI, what the tools are that they could um use to their advantage. And so it creates what this what we like to call this jagged edge um frontier, and it's it's this idea that um, you know, some AI tools are super very good at certain things. And so, you know, uh, but then it doesn't do so good at other things, and it's also context specific to, you know, potentially your company and your role. And so it feels like AI has like this very jagged edge, really good at some things, maybe not so great. And then of course it's context specific. Uh, and so if you're not helping your folks internally to understand and to experiment and to experiment out in public on within the company, uh, we're we just continue to change uh we just continue to chase this hype cycle. Uh and so it feels like there is this execution and vision gap for sure.

SPEAKER_02 11:58

And one of the things that both of you have articulated is you know the importance of the learning journey. And I often think that's underestimated in term in terms of, you know, first having your employees know how to use AI in a search capacity, something that you know most employees are already familiar with, then perhaps switching to assistants and understanding, okay, how do you how do you interact with an intelligent assistant as part of your day-to-day work? And then, you know, then the leap becomes agents. It's very difficult to go from zero to agent in a way that employees feel feel comfortable with. So I want to talk about champions, AI champions. I think this is a phrase that has become more and more common. And I think it's recognized as an important strategy, not just the top-down change, but how do you activate these AI champions and influencers within an organization? Shueta, how do you identify champions? What are you looking for in identifying these people who can help lead the charge around AI adoption?

SPEAKER_00 13:05

Yeah, of course. Um, uh, champions um at our company were not, you know, who you would expect naturally. They weren't always the most technical people. They were rather the curious people who were cross-functional generalists and who were willing to try out things, fail, and come back with a better version. So in our pilot, um, champions actually picked the participants, so they had the skin in the game. So they weren't like, you know, uh assigned a role, they chose it. So the ownership mattered. And uh after almost like six to seven months of working with these folks now, uh, in fact, longer, uh, including the pilot, it's almost been a year. A new, you would exactly know what the con what content to give someone, um, depending on where they are in the journey. So, you know, uh a new glean user always needs a different kind of a nudge than someone who's already built, say, five or six agents. So we also built infrastructure for champions to succeed. Um a private like um Slack channel where anyone could share ideas, showcase some wins, and exchange tips. Power users definitely, you know, influence other employees at the company. Um, and then we also started with um recording officers, which were not really just presentations, but they were more collaborative debugging sessions, bite-sized updates and company bulletins. And we also gamified it, spotlighting the top agent builders, for example, uh, most active chat users, most creative workflows. And that made success visible and repeatable. So our mantra uh became like bring AI to the work, don't bring employees to another tool. So embed AI where the employees actually work uh already.

SPEAKER_02 14:56

That's great. And for the folks just tuning in, we're we're chatting about AI champions and the importance of AI champions within the organization. That's something we see consistently as well, you know, mirroring that important top-down change with activating your AI influencers and champions and recognizing that they might not be who we think they are. And we consistently see they don't tend to be highly technical. They don't tend to be at the top of the org chart, but they have big, big influence in encouraging other people around them to adopt. Nicole, I'm curious from your perspective, what are the characteristics you see in these AI champions? Do they have a certain persona or footprint that tends to show up time and time again?

SPEAKER_01 15:41

Yeah, I would agree with Schwedo when she says, you know, these are usually just curious early adopters that are cross-functional generalists. That actually describes me and Helen, the co-founder, other co-founder of Women Defining AA, to a T. And, you know, just like you had mentioned, it's oftentimes they're not always the most, you know, technical, but they have the ideas. They know how their workflows and they can look at AI and say, oh, this could probably work here. And that's one of the other characteristics that we've seen time and time again with some of these champions, is that they're power users that are willing to just try, fail, talk about that failure, what worked, what didn't, and then keep going. Uh, this is very much kind of the ethos that we built our community around is just put it out there, just try, fail, fail multiple times until you can kind of find something that works. And so uh one of that's and that's actually one of the other things that we see is that they can apply solutions to different problems instead of just keeping AI to a single lane. It's like, oh no, I only use AI for this over here. No, they're they're constantly applying solutions uh to different problems. And so that's really exciting um to watch, even in my own experience, right? Uh in terms of a champion, because uh many other industries are just emerging uh with AI, and local government happens to be one of them. They're not they're they're not as forward as the private sector, but they also don't want to be left behind. And we um I'm actually leading a session later today about vibe coding in local government. You wouldn't think many local governments are are interested in that, or it's just it's pushing the boundaries a little too much, but it's speaking very much to this idea that they are the subject matter experts. They know how work flows and what if we put a tool in their hands that allowed them to create the thing that they usually have to rely on expensive consultants for or, you know, um uh, you know, other folks to bring their idea to life. And so uh it's it's great to see that happening in different sectors. And again, it always comes back to who's who's willing and who has the idea.

SPEAKER_02 17:54

And I think that's one of the most fascinating shifts we're seeing across the board is how the product development lifecycle is rapidly changing in terms of being able to rapidly prototype much easier and being able to demonstrate a proof of concept where previously you were relying on these experts. It's it's quite exciting, but it also opens up some risk for the organization and needing to be intentional about how you how you manage that as well. I love that you touched on cross-functional workers as well and those being great candidates for AI champions. It's something that we consistently see as well, you know, because in part they're not just building for themselves, they're thinking more holistically about the organization. And I think that's, you know, often a strategy we'll recommend to organizations is don't just, you know, find the people who are curious. That's a very important group, but find the people who span different functional groups and teams as well.

SPEAKER_01 18:54

Yeah.

SPEAKER_02 18:55

So I want to switch to measurement. This is very challenging in the era of AI. You know, how do you measure success? How do you measure effectiveness? Nicole, I'd like to start with you. How do you think leaders should be thinking about measurement right now?

SPEAKER_01 19:15

Yeah, you know, this has been a conversation that's happened for a couple of years now, because I think the first gut reaction is, well, how do I measure the success of this thing? Um, and especially with AI being new and new to our behaviors, folks have really tried to figure out a way to successfully measure it. And I don't, we don't know that there are really any perfect metrics yet. It's still mostly about usage, it's about building habits. Uh, you know, as we've already talked about, it's about understanding how work may shift or change. And so what good looks like actually depends on, you know, the job uh that's to be done with AI. Uh I, you know, I have the privilege of working in different industries and different sectors. And, you know, at first it was always about the productivity metric. It's like, yeah, but then if you are just trying to get so many hours back there, then, and if you just take on more administrative work, is that really helping the organization? And that is something that I've seen echoed across multiple industries. And so then it becomes about, okay, well, what's those higher order goals that we could be uh going toward? Uh, whether that's um, you know, a higher an overarching uh higher goal of, you know, in local government, it might be something like um, you know, maybe citizen satisfaction or something like that. Um, but again, like folks are still trying to figure out like how that attaches, like how does it all connect? So it really for me, it's it's it's been very contextual. It it really depends. Function to function.

SPEAKER_02 20:51

We see team to team anchoring in those existing business KPIs, I think is is a hallmark of what we see work given how malleable and and flexible the the technology is. Now, closely related to the conversation around metrics is this phenomenon that's sometimes referred to as AI theater, you know, using the technology to use the technology, there's a lot of pressure right now to show you're using AI, show you're investing in AI and you know, building that capability within your organization. Nicole, how do you see AI theater show up day to day? What are some of those telltale signs?

SPEAKER_01 21:31

So um I've said that I get to see across industries. And it's always when I hear that somebody's, oh, we're using Chat GPT um on like a superficial level. And the the real kicker for me is when they say that, oh, it's the free version. And I'm just like, okay, that means that it's just it's just a lot of hand wavy, it's a lot of theater. Um, you you don't really have that grounded strategy. Uh And like I said, it pops up everywhere. And so I think externally, CEOs make a lot of big statements about AI. But employees don't actually know what it means for their day-to-day. And they're not prioritizing the training or the experimentation in public. And they can just say, oh, well, we're using Chat GPT. But again, like all the things that we've talked about, whether that's the training, understanding how it works in the workflows, are we failing at it? We want to hear like those failing stories just as much as the success stories. And so, you know, if folks don't have that guidance and it really is just a bunch of hand waving, internally, like people just lack the sense of proficiency. They don't know what good AI looks like for their role, for their team's role. And that's just where it's very frustrating to hear companies still engaging in that way, uh, just so they can say they're doing AI.

SPEAKER_02 22:57

And I think one way to combat this is Schwetza, what you were discussing earlier in terms of the importance of embedding AI in the flow of work. And I want to return to that conversation because I think it's it is very relevant here. Can you walk us through one workflow where AI went from a side project, maybe more of that theater and talk as opposed to action, to it being deeply embedded in the flow of work, in the day-to-day work of employees?

SPEAKER_00 23:31

Yeah, of course. Um the way I kind of think about is uh our agents or our automations that we've built with AI or Glean so far, uh, they fall under like two buckets. One um is like evergreen or always on kind of agents. And then the second bucket, I was I would call it as seasonal. And both are essential. Um evergreen, always on kind of agents would be answering the same questions that never go away. And the and a good example here is, for example, like the uh IT help agent is the is a you know, is a highly uh used agent in our uh company. We already had a Slack channel where employees asked IT questions, right? The problem was not that the answer, the problem was actually here that the answers were buried in the scroll history and not discoverable in that moment when people are looking for. So every question felt new. And obviously, there was earlier a human interacting or answering these questions. However, we put a glean agent in that existing channel, and now it's the first-line answer. It deflects repeated questions, pulls from historical, you know, um databases, docs, and routes, only the genuinely new issues to humans. So we didn't create a new process here, we made the existing ones smarter. And the same uh principle kind of applies to another agent that we are uh soon rolling out, instead of um, you know, pinging a data analyst, and this is called a data help agent, where you know employees will now be able to ask natural language questions and get answers pulled directly from Databricks environment in our case. And that's you know a saving for both the people asking and also the team that used to field those requests earlier. So these are some of a couple of examples of evergreen or always on kind of agents. Then there are these seasonal agents that we experimented with recently, and uh, you know, because uh these use cases are kind of built in, they have deadlines and captive audience. Open enrollment was was a perfect example here. Every year, people team gets flooded with the same questions about plans, options, deadlines, workday actions, etc. So our um uh, you know, the benefits agent named to feel approachable, um, or you know, not just um employees get clarity instantly and get pointed answers right away instead of waiting for an HR response. And another uh seasonal sort of an example is a self-performance review agent, which we launched ahead of the review cycle, who likes to write their self-reviews uh every quarter or every six months. Employees could only ask the agent to structure their assessment, provide examples, and you know, how to and get help on how to frame the impact right at the moment they need it. So that those were some of the impactful agents that we recently uh you know started using. And they've shown impact um significantly.

SPEAKER_02 26:44

That's great. And you know, there's a there's a comment in the in the chat from from Joshua that I think speaks to this really well as well. You know, we often think that AI agents should do everything, they should automate the complete process. And what you've done so well is you've thought very carefully about the division of labor. You know, for the aspects of work, for example, that are routine and there's precedent for, you know, delegate that to an agent. The the novelty in terms of new tasks, new, new, you know, asks from employees, that's you know, a uniquely human pursuit often. And I think that division of labor, it's not easy, but I think it's important when we think about agents in particular and workflows. So I wanna I want to wrap up with one final question. And uh Shueta, I'll I'll ask you this one first. You know, there's a lot of pressure to move quickly with AI now. And I think there's a lot of fear of going too slow and feeling behind. What is one concrete action you'd recommend folks take in the next week or next month in terms of taking that extra step to move their AI adoption forward?

SPEAKER_00 27:58

Yeah, if there's one thing I want um I want you to walk away with today is it's not uh basically it's don't start with a strategy deck. Start with one flow, workflow your team already does and make it smarter. That's it. That's the whole playbook. Um find the friction, drop AI in it, pick one metric to prove what's working, and then let your teams pull it forward. That's how we went from you know a 50%, uh 50 person pilot to uh 700 plus agents, not by planning it all upfront, but by following the signals one workflow at a time. So we need to divide and conquer and you know, um, not everything can be implemented on day one. So I promise you, once your team feels that you know they get 20 minutes back, the next question um would be: should we use AI or should we, you know, um where where can we use AI um better? So basically embed the AI in the processes themselves that already exist.

SPEAKER_02 29:00

That's great, great advice. Nicole, how about you? What's one action that you'd recommend every organization take as part of their AI adoption journey right now?

SPEAKER_01 29:10

Yeah, so uh as I as I mentioned, I have my own AI company and I am typically because I'm this, I'm essentially the customer um and the subject matter expert, I oftentimes will prototype out new feature sets and then hand off to my engineering team to put into production. And um, I tell the story as an example for like the takeaway. Uh I was prototyping basically an ADU uh analyzer. So ADUs are accessory dwelling units. Uh, they're a great way to jumpstart affordable housing initiatives within uh jurisdictions. And as I started to build it, I, you know, is like eh, killed that uh you know session, started up another one, some vibe coding, and tried multiple ways on how to solve it. And ultimately what came out of it was a new idea that I hadn't even thought of beforehand. Because after I kind of built a prototype of where could these detached accessory dwelling units go in one of my uh trial uh municipalities or cities. And then I realized, you know, what would actually make this great is if you could do a policy analyzer on this and basically adjust policy metrics. So, hey, what if we all the reduce the residential uh footprint for these parcels to 5,000 and we made it so that any ADU could be on a 5,000 uh square foot parcel because every jurisdiction is different, and started to do some toggles there. And so I was like, oh my goodness, that's that's actually the coolest part. And that's not what I intended or started with. So my takeaway for you is to take one recurring problem and solve it six different ways with AI as a team exercise. Um, hopefully you'll have some of those failure points because maybe you know, one, two, and three didn't work out well. But maybe by the time you get to a sixth version, there's something interesting, and maybe something interesting came out of it that you didn't initially intend it. So six different ways to solve one problem and align on the best pattern and make that a habit.

SPEAKER_02 31:12

I love that. And and so concrete and it aligns with, you know, overwhelmingly, there's evidence to suggest that we're more creative, we're more innovative when we have a greater quantity of ideas to work with as well. And I think that very much feeds into that. So six different ways. I think that's something we can all put into practice. Well, thank you, Shwata. Thank you, Nicole, for joining us. Thank you, everyone, for tuning in. I learned a lot. I think our audience did as well. And uh stay tuned for for the next episode. If you enjoyed this episode, subscribe to Intelligence, Real and Imagined so you never miss a conversation. And download the AI Transformation 100 report using the link in the show notes. Thanks for listening, and we'll see you next time.