What Do Your Colleagues Do All Day? (The Value of Institutional Knowledge & AI for Process Reengineering) with Jennifer Smith, CEO at Scribe

o you bring the general intelligence, what's coming out of the labs and the models, to bear in the specific operating context of a firm?

And this is the big challenge. This is I think m- the crux of why we're not seeing more. The, the underlying models are now incredibly intelligent, right? They're PhD level intelligence, but they have been trained in the generic ways of doing things. These labs have spent literally billions of dollars training models on how to lawyer and accountant and investment bank and all of these things.

But they don't know how you do that thing here at Acme Corp, right? And the challenge is Acme Corp can't tell you how Acme Corp does those things in any kind of scale at any level of detail, right? That's a really important asset of the company. I'd argue it's its core IP, but it's pretty ephemeral. It's in people's heads, it's in their expertise.

Maybe it's documented in scattered places correctly or incorrectly. The big challenge is how do you then take the special sauce of what the company knows how to do and then make that legible, not just to its people, but machine readable to its agents? And so then you have agents plugging into the context of the way that the company is operating instead of today we have basically said, "Here's a Nobel laureate, do amazing things with it."

And I make the analogy, I feel like most enterprise AI today is I hired the really expensive Nobel laureate. I've spent a lot of money on it, by the way, but but they're sitting in my lobby and, now six months have gone by and I'm asking the question, "Why am I paying so much? I haven't seen anything out of it?"

It's have you invited the Nobel laureate in? Do they have a badge? Do they know what they're supposed to do? Have they talked to the people? Have they shadowed? Do they understand what makes your company special and do they know where they should be plugging it? And I think that's the big challenge that we now need to overcome which is as much a technology problem as it is a, a people processes change management problem.

Richie Cotton: I love that analogy with AI as a Nobel laureate, like it's a vanity hire. Yeah. Okay. We've got lots of, 

Jennifer Smith: Who do great things. Like they have great potential and they have great intelligence, but intelligence alone Does not solve your problems. You have to apply it. It has to know what problems need to be solved.

Richie Cotton: Absolutely. I feel like there's a, a lot of challenges in making this work. What's step one then in getting to be able to apply all your AI, getting to apply all your intelligence and making it more specific to your company? 

Jennifer Smith: I call this specialized intelligence. It's not a new concept.

It's this idea of all of the kinda special sauce of a particular company. Now, your company's been around for years, maybe decades, and you have learned how to do things a very special, particular way. That's your competitive advantage. It's the set of decisions, accumulated set of decisions and judgments and values and exceptions, and all of these things that collectively have come together so that you know how to uniquely service customers, ship a product, manage exceptions, all of these things that you have to do.

And again, the challenge for most companies is that core IP is pretty ephemeral. It's mostly scattered in the expertise of people's heads. Maybe a small amount of it is written in places. And when your company was all just people people kinda paper over that. If you spend any time in an operations floor, I've spent a lot of time in operations floors actually, and I sometimes have done this as a fun game.

You sit and you look at your c- your watch and you say, "How long until I see somebody pop their head up and tap someone on the shoulder, proverbial or literally, to ask them to show them how to do something or explain something to them?" And it's usually a few minutes if you're, like, in a big enough center.

Humans do this all the time. They kinda figure it out. It doesn't work that way with agents, right? And so the challenge then becomes how do you harness this specialized intelligence, actually create it so that it's an asset that you own that's both legible to people and legible to your agents? And the thing that I would posit that I think most leaders now need to think about is general intelligence is meaning the LLMs, the models, the generalized intelligence, that's just something you rent, okay?

That's not your competitive advantage. That's not a durable moat. That's not a strategy. It is like electricity, okay? It's incredibly valuable. It's actually critical to being able to do business. The invention of electricity changed the arc of human innovation and society at large, right? I think, AI and general intelligence will do the same thing.

But you don't win because you have better electricity than the guy next to you, right? You win because you built a better factory, and that factory knows how to do specific things, and that factory gets better and better every time it does something. And so you need to think of your operations more as, okay, I rent this incredible intelligence from the labs, from open source models.

I think you'll increasingly some- see some more of that over time. And really my specialized intelligence, how do I harness the things that are special and unique to me and make that an asset I actually own rather than it living in people's heads and being ephemeral? It's an asset that I actually own, and then it compounds.

Every time that a person does work, an agent does work, we are compounding in that understanding and we're getting better and better at what we do. And I think it's a very different way of thinking about how you operate, but I think that's what's critical if you've, if you are at all interested in figuring out what's my competitive advantage in an AI first world.

Richie Cotton: I love that idea that you've got these ephemeral processes that are in someone's head. I'm just thinking ooh, how does a- DataFrame podcast episode occur, and actually a lot of the processes are in my head. So is the secret to just write stuff down, do more documentation? 

Jennifer Smith: That is the brute force method that we are seeing a lot of companies try to do today.

I talk to leaders every day, and I say "Okay, like we think this is the imperative. How do you think about this?" And we're increasingly hear them saying, "We agree. We've come to similar conclusions." And so what we've started to do is map out what all of our processes are right now. We've tr- created some kind of top-down exercise, and so that looks like everything from, "I've heard we hired a bunch of consultants, and we locked 500 people in the basement ballroom of the local Marriott, and they interviewed them, and they like wrote everything on the whiteboard," to we do ethnographic studies where we like, look over people's shoulders and shadow them as they move throughout a hotel to figure out what our operations are there to, we hire a whole bunch of consultants.

By the way, I was at McKinsey for 10 years. I did a lot of this work, right? To kinda like to map it for us. I think you're increasingly seeing that wrapped up in a lot of the FDE, forward deployed engineer work that is being done. A big part of what the forward deployed engineers are doing is trying to figure out what the heck does this company actually do.

What are the processes, and then where do we see opportunities to build agents and to automate it? I think this is incredibly ironic. Like we have s- we're seeing the greatest, intelligence or at least artificial intelligence that, that humanity has ever seen, and we're just throwing bodies at the problem to try to do like manual mapping and interviewing and shoulder tapping to understand how work is done.

The good news is we can... This is actually a problem LLMs are excellent at, right? We can use them to passively observe and understand how work is done, and take all of the raw, unstructured inputs of what are all of the work activities happening here, and actually structure and classify it in a way that is readable both to humans and to agents.

And I think that's the critical foundation for anyone who's trying to think about how to transform their business, is first answering the question: what do we do today? Because if you wanna be better tomorrow, you've gotta figure out like what's the work to be done? And then you can't transform the whole business all at once, and so what's the work to be done, and then where are the areas where we think we're gonna get the most, juice from the squeeze, and like what do we think that looks like, and let's build the business case and build a plan to go do it A- and we meet a lot of folks who have shortcutted different steps of that process, and often to, results that were not what they had hoped for, and then they kinda come back to the beginning and say "Oh, wait a minute, we need to build the foundation."

Richie Cotton: That is interesting. I think nobody really loves writing documentation, and having a consultant stand over your shoulder and write documentation about what you're doing, that just seems slightly creepy and a, maybe a little bit weird. 

Jennifer Smith: Yeah. Yeah. Fun, fun to be the consultant. I'll say it's not fun for the consultant either.

Nobody wants to be doing that exercise. Y- I started this company seven years ago at Scribe, and our first product was one that automatically generated SOPs from watching you work. I would get email, I still do to this day, I would get emails or DMs, I've gotten Christmas cards from users who would say "Hey, you don't know me, but I just wanted to say thank you because I previously was creating this documentation, I was, like, copy-pasting screenshots and writing things down and all, you know, process mapping on a whiteboard, all these things. And that is really manual, dare I say soul-crushing work. And thank you for automating that away." And look, I am in no way advocating that we return to that world where people are spending all of this time because it's one, it's not fun, two, I think it's a terrible use of human time, and three, what you get out of it represents just a single moment in time, and it was like the way this thing was done this one time when I interviewed this one person, and processes are incredibly dynamic, and by the way, they're highly variable, right? So whatever that one sample size, you pick that one person out of that team of 20 I guarantee you those 20 people are not all doing it that same way, right?

And so it's not even accurate. 

Richie Cotton: All kinds of chaos there. I don't if- You, if you've got a website, you can see where people are clicking, and you figure out a customer journey from that. Is this is this automated technology approach, is it something like that? Like, how do you track what a process actually is automatically?

Jennifer Smith: Yeah. It's a it's a similar idea, but it's now saying across all of the business applications that you use in a company, across all of the teams, what is the work activity being done? And it's taking all of the raw data, similar to a website tracker it's taking all of that raw data and it's parsing it into workflows.

This is a big part of what at least we have spent a lot of time understanding because enterprise workflows are a gnarly beast and that's obvious to anyone who's ever spent any time working inside any of these companies. But being able to tie these things together and say, "Okay. This looks like this.

These tasks are all similar tasks. They're just variations of the same theme. And then by the way, these 10 tasks are all part of this bigger workflow, and here's how we collect and all aggregate it together." The good news is this is something that LLMs are quite exceptional at actually, right?

This is something that previously never would have been possible because we're talking about amount of data that is just so much bigger than any human or any system previously could have been able to parse. But now LLMs are actually quite excellent at it with the right training and kind of understanding.

And so we're now able to map, with full confidence, what are all... what's all of the work that is done at this company? What are all of the workflows that we do? Let's in as much detail as you want or as high of an abstraction as you want, map out a very detailed from state of like this is everything that's happening today at a high level down to all of the different variations.

And then once you can have that, you can use that in a variety of different ways. Most relevantly for people today, most of the conversations I have is to drive enterprise AI transformation. Okay, help us understand where we can deploy AI, what it's actually worth to us, what it would look like, and then give the right instructions that we need so that agents are doing the correct thing.

Richie Cotton: Can you give me an example then of a process that's been discovered using this this technology, and then the company's gone, "Oh-" It's obvious we should make a change here, and something good's happened 

Jennifer Smith: There are so many, 'cause I would say like everyone we meet says when the, certainly when they're talking about sort of a process level, they're like, "I know this isn't the way that it should be."

And then they see the data coming out of it, and typically their response is this is even worse than I had imagined it to be" of just all the different variations within that. It works best when you have a point of view on what it is you're trying to drive, right?

So we meet folks in who will say, "I have a mandate to do more with less and to use AI to do it." And I say, "Okay, great." That's like everybody has some mandate that's some flavor of that. What's yours specifically? Or what are you trying to drive? And so then you get down to actually, we need to ship products much faster in market because it's changing so dynamically, we need to be faster," right?

Or, "We have a $50 million cost number that we have to bring out. Help us understand what we're being really inefficient." Or, "We have to ramp our sales reps faster. We have to get, better sales productivity." And so if you just take that one as an example, we worked with a fairly large company, a Fortune 1000 company.

They mapped this detailed from state for their sales team, and they saw that they were spending, about 50% of their time on non-selling activity, right? So you had your average rep, maybe they spend 30% of their time on the phone actually with customers, and then, they spend probably an additional about 20% doing related tasks, prospecting, following up on POC, all that kind of stuff.

And so they said, "Wow, 50% is a really high number," right? How do we reduce non-selling time so that you can get just more quota productivity per rep, right? And so they were able to look at it and see, okay, there's big categories of time that we're seeing across reps that we can pull away from those reps, right?

So one category was actually enablement was not very good. They were doing the sets of activities, but it wasn't landing with the reps, and so they were spending a lot of time doing their own enablement, and they surfaced this. And they said, "Okay that's b- that's a bad use of rep time."

All right we see all these things are gonna... We're gonna s- we're gonna pull that into enablement and make sure we actually cover that. There was a whole other category which was reps were spending a whole bunch of time answering security questions. The company obviously had a dedicated security team but the clear roles and responsibilities were not defined, and so there was a whole bunch of work that they were doing.

They said, "Okay, actually, we could automate quite a bit of this through a much more robust agentic security answering system where you generate a bunch of the answers for the rep so they don't have to go search." Okay, we can reduce a bunch of that time. There was a whole other category around similar legal back and forth.

And so anyways, they were able to find all of these sort of places where gosh, they're spending a bunch of time where they shouldn't, and so how do we pull that away so they can be spending more time in front of customers and we can see more quota productivity? It's similar to that was just a, and this is, anyone who's ever run a sales team before can relate to this.

You have varying performance across your across your reps. That's true of any team in sales. It's very easy to measure. And so you say, "Look, what are my best people doing that's different from, my other people, from the rest of the team?" And so one of the interesting things has been able to look at some of the variation and saying what are some of the best practices that the team's already doing today?"

We're not even talking about doing something completely new. You could even just look at, what are some of the things that our best people are doing today, and then how do we replicate that across the team really quickly? 

Richie Cotton: Okay. So it seems like you can get to real deep business insights there where you say, "Okay, this is where people are wasting their time.

This is the thing that's not effective. Let's go and change that." 

Jennifer Smith: Or this is something that's actually quite effective, but is really unevenly distributed. How do we identify that and push that across everybody? 

Richie Cotton: Interesting. Are there any things you've seen across many businesses where you think most businesses do this thing really badly?

Have you got any insights into what is typically done wrong? 

Jennifer Smith: There is massive variation in nearly every workflow that w- that is done by any reasonable number of people, any kind of, with any kind of reasonable repetition. And the first reaction might be to say "Gosh, that's bad.

We should just, variation is bad. We should stamp it out and we should standardize it." I actually look at that and say, "Hey, that's really interesting." It's like you're running an experiment, and there are some better ways of working than others, right? There are just some better ways to get a workflow done where it's more efficient and you have higher accuracy or whatever it is that you're trying to go after.

And so you should look at that and then actually understand what was different about the thing that's really great, and then how do I push that across everyone? C- 'cause I'd say one of the common things that when we meet folks, they immediately wanna get to "Okay, I wanna agentify everything, and I wanna fully transform my business with agents."

And that's great. You can do that. We can help you with that. That can take some time. There's often a lot of really low-hanging fruit that is just you know what? Your one person has found a much better way of working, and it's like it already exists within your company, and we just need to diffuse that knowledge to everybody else.

That may not be the sexiest thing, but that guarantee you, that exists within Pretty much every workflow within every business that we have seen, and that's like a relatively quick win, right? And maybe part of my message is there's a lot of quick wins along the way to the road of, adding agents to the team and driving, much broader AI transformation.

Richie Cotton: I love it. It's just surfacing all the kind of the, the stupid wasteful bits of of work and saying, "Okay, maybe let's not do that anymore. Let's do the cool stuff that we found instead." 

Jennifer Smith: There's a lot of stuff like hey, we bought this tool for people 'cause they were swivel chairing between a bunch of different tools.

They were doing this thing manually in Excel. And I can't tell you the number of times that I've sat down with a C-level leader who's like, we bought this tool for them. They solved it." And then we show them the data and they're like Gosh darn it, they're not using the tool. Like maybe they log in, they use it for this one little part, so the data looks good, but then when I actually w- use look at the workflows, they're not using it for the workflow we had intended, right?

And so you're like, okay, why not? Why are we still doing things in a really, kind of manual, outdated way? Gosh darn it, we look at it and people are copy pasting information between systems. We should just build a connector between these systems, or like they're working around this tool because it just, it needs to be configured better.

Again, these are things that are not sexy. They're not, I would say they're not sufficient to drive transformation, but they are pretty easy, quick wins that when you have the data becomes obvious very fast, and you can say, okay, like these are ways that I can free up some capacity along the way. 

Richie Cotton: And yeah, I can certainly imagine how that's gotta be quite good for employee morale as well when you're finding these process problems and suddenly, oh, that you don't have to copy and paste stuff.

The tools just work for you. 

Jennifer Smith: Yeah. I'll give you an example just e- even from, running this within Scribe, like reps are, or our sales reps always complain about Salesforce and different tools and, whatever, and it's like really hard to separate the signal from the noise.

And then, we like took a look at the data, and it was like obvious immediately that there were a bunch of fixes that we could make within a day in the way that we had configured Salesforce, and we had set up some workflows that radically improved their daily experience. And so they love running our product because they're like, "It just makes our lives better."

It's all the things that we have complained about, and maybe some things that didn't even hit the level of complaining, but were always a nuisance, and now that gets flagged to people who can actually make the changes really quickly. 

Richie Cotton: Okay. I love that. So I'm curious whether, do you have to map everything at once?

Is this kind of like you, you get the whole company's processes for everything, or is this something that you can do inc- incrementally? If you do it, where would you start? 

Jennifer Smith: So I would think of this as a com... If you think about harnessing, owning, and compounding your specialized intelligence, right?

That's something that you wanna do on an ongoing basis. If you are a dynamic company, you are gonna be continually changing what you are doing, the expression of your specialized intelligence. Increasingly, I think the boundary of the work that humans do alongside agents is going to change. And so I think the best companies are those that are gonna be able to adapt to that fastest and get tighter and tighter in those learning loops and in how they compound- That specialized intelligence.

And so this is not a one and done exercise where you look at it and you say... this is what I used to do as a consultant. It was a one and done exercise, right? And it was actually great for us 'cause we'd sell you a three-month engagement, and then we'd come back a year later and do the same thing again.

But ideally, you're doing this in a continuous, ongoing way. And so our recommendation is just to say, "Hey what's the top priority for the company? And let's focus on that." I never wanna do AI transformation for the sake of AI transformation. If your answer is, "Because the board told me to," or, "Because our CEO told me that we have to do AI", I- that's, that, that's not a good answer.

We're not gonna get to success there, right? And I kinda always push people and say, "But why? But what are you trying to drive?" And usually there's like a good answer underneath, right? Which is, "We are trying to scale up. We can't hire fast enough." I was talking to someone in the construction industries.

Our business is booming, and we can't hire fast enough, so we just have to figure out how we can handle more. We're work-- It was interesting. We're working with a European company, and they want to d- move to a four-day workweek. Do all the same work, but be able to do it in four days for their employees to have a three-day weekend.

And so very European. And so you're like, okay, there's a, there's like a clear why, right? So what is it that you're trying to drive? Let's start with what you're trying to drive. And then you can look at it and say, okay what parts of the business are in scope versus not? For things that are, company-wide, like we are trying to drive broad efficiency or, we are trying to figure out, How we ship products faster.

That tends to encompass quite a bit, and so we would say you wanna capture all of those workflows and you wanna have a snapshot across the company. If your answer is, "We wanna drive rep productivity," then I'm gonna say, "Okay, we should only look within, kind of the sales organization."

But what you wanna do is get a clear view of what are all of the activities that are being done here, 'cause then very quickly you start to surface, and then here are some of the biggest issues or opportunity areas that you can then go focus on. 

Richie Cotton: I say that four-hour work week sorry, a four-day work week does sound pretty amazing.

And four hours actually would be even better. You mentioned the construction industry, and I'm curious as to like how you go about tracking things automatically there, 'cause it feels like construction is very manual tasks. There are things that aren't just in digital systems, there's people walking around a construction site carrying tools and things. How do you track all that kind of stuff? 

Jennifer Smith: Yeah, look, there are companies that are focused on the physical world, right? We think just about the digital world, so I can only speak to that. But we work with a lot of industries that have, big physical components.

So think construction, think retail, think hospitality, think manufacturing. When you picture that in your head, you probably picture like a, a guy in a hard hat, on a construction site or in a, in a big factory or something. There's still a ton of digital computer-based work that happens in these businesses that often, touch those processes.

And for example, I was talking to someone in agriculture and they're trying to figure out how do they get higher crop yields, right? And you would say, "Okay, that's like mostly has to do with the farm-" actually, there's like a ton of planning, that goes into the whole process.

Everything from when do the trucks arrive, to the refrigeration and the supply chain all these sets of interdependencies that are still quite digitally focused. And so there's still a lot you can get even from big businesses that have, a big physical labor component. 

Richie Cotton: Okay, yeah, it's pretty rare now where you've got like businesses where there's still pen and paper everywhere and s- 

Jennifer Smith: no, but, you even think of like a, a hotel, for example, right?

So we're working with a a big hospitality company, and they're trying to figure out how to operate their hotel franchises more efficiently. And you think, okay, I don't know, the bartender, the front desk, like how much is there? There's actually a ton, right? Before we met them, they were just sending people to go and shadow the person at the front desk and follow them around.

But like they're checking you in. You have events people who are coordinating. You've got folks who are doing the ordering. There's all of this digital work that goes to support it behind the scenes. 

Richie Cotton: No that, that's very true. Even if you've got a, a physical world action, then there's probably some sort of digital equivalent backing it up.

Or at least a, a paper trail there. One of the things that surprised me just even in the very early days of Scribe is, one of our earlier customers was a trucking company And we were like, "Okay, trucking is not this is not an obvious..." We do a lot of financial services, healthcare, what you think of traditional knowledge worker industries.

Jennifer Smith: And they were like our truckers still have to interact with software. They're signing in, they're signing out, they're handling deliveries. We have still to do enablement to make sure that, people know what they're supposed to be doing, and that they're using these systems appropriately, and that we're providing them the right systems so that they can be doing their jobs when they're in the field."

Richie Cotton: It seems like there's a bit of a pattern in that a lot of the businesses you mentioned, the examples you mentioned, there seems to be, they're cases where you've got lots of people doing similar sort of things over and over again. Are those the kind of processes you want to optimize then? The stuff that I guess gets done a lot?

Jennifer Smith: Yes, certainly, if you talk about like the one-off special snowflake generative work, that- that's probably not a prime example for what we're talking about here. For example, my job. If you were to try to build a process map of what I do I'm an N of 1 at the company.

Nobody's job looks like mine. My job looks pretty different month to month, even day to day. I'm quite scattered in different things. I- there's no... You won't find repeatable p- processes, right? But if you go into k- any company and you look at the FP&A process, for example I would hope that there's actually, there's there's some method to the madness there, right?

There, there is some logic of process where you've got a set of folks who are trying to execute, certain tasks, and it starts to look similar. Most companies of any kind of scale are built on the back of repeatability of some of these processes. Now, hopefully you're pushing the boundaries, you're continuing to innovate.

They look different over time. But you're doing something, again and again to to be able to deliver some kind of cus- customer value. 

Richie Cotton: Yeah. Yeah I hope the the finance team, like the accountants aren't making up new processes every month when they're trying to figure, like, how much money do you have in your business.

Yeah, that ought to be repeatable. 

Jennifer Smith: Mom fries in a garage absolutely, that this is not for them. But it, if you've been running a a financial services company for a decade absolutely. 

Richie Cotton: Yeah. Okay. Suppose you've figured out all you- you've documented what your processes are.

You've got this process map. How do you go about executing on the change then? Is this, does this come from like the C-suite down? Does everyone have to look at their own processes and go, "I'm gonna change this"? Like, how does it work? 

Jennifer Smith: This is what's been really fascinating to me is to see how companies right now are grappling with, and I'll just latch onto the AI specific component of it right now, 'cause you can drive transformation that doesn't involve using agents, right?

I think this is one of the things that I see a lot of our customers grappling with right now, which is in 2025, certainly the first half of 2025, I would say a lot of companies' AI strategy was what I'd call let a thousand flowers bloom, where they basically looked at every, all the people and they said, "Employees, your job is to figure out how to use the AI.

You have an AI mandate, go buy the tools, and our hope is that something really interesting bubbles up out of this." And I think we realized pretty quickly that's a, that's like a pretty unreasonable, and I would argue unfair expectation for most employees. And so you got some interesting things out of it, but you no- you don't get, enterprise-wide transformation Right?

And then you have folks trying to come top-down and say, "Okay, now we're gonna do this concerted effort from the top, and we're gonna push it down." A- and then what's the failure mode there is we didn't actually understand or appreciate all of the different ways that work was done, and so you try to get this thing and it doesn't accurately reflect, and so a whole bunch of stuff breaks when you actually go to put it in production.

It's like for anyone who lived through the RPA era a lot of the problems that existed with RPA, where you have some kind of centralized team that's like, "Aha, we will build a bot to do the things that you guys do over here," and they built it for the happy path, and they built it for the eight-step process, but the reality is it's a 12-step process, and the happy path only happens 10% of the time, and there are 20 other variations that they just didn't understand, right?

And so I think the right answer is actually to have a marrying of the two, certainly where folks feel empowered for where they can see, parts of their own workflows or their team's workflows where they can leverage AI, and I've seen so many cool examples. We could talk more about that.

But where you're also marrying that with some kind of top-down perspective on what we're trying to drive. Again, not we need to use more AI, but we need to, find 50 million in cost savings, or we need to ship products faster, or whatever the business mandate is that you're trying to do. And then from there you can identify here are the top 10 places, within this department that we think we could get the most out of it.

And okay, now let's go look and see within those areas what would a from to look like? What are the options? And then you can go engage with those folks on the ground, and you can have a, a coming together of the two. And this is very much how we're building towards, 'cause today a lot of that coming together is happening still with people having conversations.

And once you understand, again, you've harnessed all the specialized intelligence of a company, you understand all the way it works, you can actually do more of that automatically, right? And then you can do that at scale rather than through some kind of concerted, center of excellence that's driving a change management program.

The thing I get really excited about is, like, how do we bring those things together through technology rather than today, folks are trying to stitch it together. 

Richie Cotton: Yeah, it just seemed like one extreme where you've just got individuals trying to figure out how to rework their own jobs, and the other extreme where you got this CEO or the board members going, "You must change everything."

The, neither of those extremes are gonna work by itself, so you need to have some kind of a combination of the two. 

Jennifer Smith: Exactly. And I-- and ideally what you have is, "Hey, here are the goals of what the board and CEO is trying to drive," and, "Hey, as you're going to do your work, Jennifer, hey, there is actually a-" a set of agents or skills that your company has created that are relevant to what you're about to do.

Do you wanna use these? Or, Jennifer and her team can identify an opportunity, create it, and then publish a set of skills that are available not just to her team, but to a variety of teams across the company that may face similar problems. I think you're seeing in the, the bottoms up, let 1,000 flowers bloom, there are many problems to that.

One of them is just duplication. Like, when everybody has created an agent to manage their calendar or something, like that may feel good, but I think folks have already quickly realized That's not innovation, that's actually just wasted effort, and that should've been something that was then published, in some kind of marketplace or library across teams.

Richie Cotton: Oh, yeah. So managing skills, that's a, a pretty hot topic at the moment. I think that is the case in many companies is like everyone's gone, "Oh yeah, I've just got Claude and I'm gonna make some skills," and just making those discoverable and shareable and, Yeah, talk me through, like, how you approach, like, managing all these things that people are making.

Jennifer Smith: It's hard, and I don't, I have not seen anybody crack the nut on it. I'll share a bit of how we've done it that has things that have worked well and haven't worked well, and then how we think about productizing that for, for others. What's worked really well for us is folks being vocal and sharing examples of what has worked for them.

We kinda have folks within each team who are maybe the biggest AI evangelists of that team, right? And so rather than saying, "Everybody needs... accounting team, all of you now need to be really good at accounting and building agents," we instead say "Oh there's one person on the accounting team who's really forward-thinking about agents," right?

And so we say, "Okay, great. We've got some other resources in the company who can help you think about how to build some of these things, and and you can build on behalf of the team and then share some of those best practices." And we have formal mechanisms for that through Slack channels, all-hand meetings, team meetings where we'll spotlight some of those examples.

The failure mode, and that's coming from some failure modes for us where, we have, particularly in our company, a lot of pretty AI native enterprising people. And we had sales reps who were like, "I'm gonna build all my own dashboards." And at first you're like, "Wow, this is so incredible that you built all of this stuff."

But then when your fifth rep has built another dashboard and spent eight hours building all of these S- you're like, wait a minute, this doesn't actually make sense anymore, right? We need to pull this back, and RevOps owned it and said, "Okay. It's really helpful to know what you wanted to build for yourself, the problems as you saw it.

By the way, we're gonna combine that with our understanding of not just your specific problems, but the entire sales force's problems that may look a little bit different than yours." And "We're gonna create these tools and push them out, across the team for you guys to use." And so again, I think it's a bit of that, going back and forth.

We think about how, again, we can build this in product for us, 'cause we sit in a unique place where, you know, having all of the context both of what you're trying to drive across the company, but also how each individual person and team works, we're able to actually marry them together and say, "Okay, to try to drive what we're trying to do at a at a, a company level or department level, we're really interested in driving rep efficiency, for example, reducing, time you're not spending selling."

We can then go and create a bunch of skills or agents that RevOps can then approve, that can then get pushed to those AEs when they're going to do work. So it's like, "Hey, you're about to do this, you're about to do like a, a Salesforce process for upsell. We found a way, we're just gonna reduce this dramatically.

Click here and, it'll kinda do it for you, and it'll prompt what it needs from you. And so I think the future is gonna be these things actually coming together and having that all done by, by an intelligence rather than today humans have to communicate amongst themselves and adjudicate how much of this is gonna come top-down versus bottom-up and meet in the middle.

Richie Cotton: Okay, yeah. It does seem like there's a big opportunity there for anyone who can crack this. But I guess, yeah, the short-term thing is you just gotta speak to your colleagues and see what they're doing, and it is useful. Okay. I guess more generally, is there a way of tracking any progress on how your processes are changing?

So you talked about okay we've got this process map, but it should be a continuous thing updating this. How do you even track whether you're getting somewhere? 

Jennifer Smith: Yeah, I think this is really important for, especially for anybody who's interested in ROI, and I would say a year ago not that many people were interested in ROI.

And like thankfully now I think almost everybody is very focused on ROI, whether they, just got there through the natural evolution, or they got really burned by a large token bill or something along the way. Step one is let's actually have projections going into this effort of what this is worth to us.

And I know that may sound simple, but that, that's often pretty hard to do, and, I think a lot of folks skip it and will say I don't know. It seems like this is roughly right." And I can't tell you the number of folks that I've talked to who said, "We spent $20 million to automate a process that cost us $10 million at the end of it," right?

Or, "We built this thing, and people didn't actually end up using it because, again, it only worked for this small subset," or the quality. We built something that was 80% accurate when humans did the process, and now it's 50% with agents, and so we're defaulting back to the... So there's a lot of kind of failure scenarios within there.

And so step one is first of all just let's build a business case, and the reason business cases have been hard historically is getting your arms around, like, how often is this thing done, by whom, what are all of the variations? What is it actually worth to us? But when you have all of that data, it actually becomes pretty trivial to then say, "Okay, we think this is the from state and this is the to state.

This is what it would actually-" This is the juice that we anticipate we can get out of it, and then you can make the decision of go, no go. Is it worth it to do it? And then ideally, you wanna be able to close the loop after you, you institute the change and say "Did we actually make the change, and did it result in what we had hoped for?"

If you take just my example of folks realizing that their employees were not using the tools they bought them, like the ROI they gave to the CFO was, "Oh look, we bought this tool, and it's gonna reduce the amount of time." Like the one example I'm sharing had to do with ad buying, right?

It's gonna make us more accurate and faster in our ad buying, and it means that like our team can handle more, and like it looks really good on paper. And then you look at the reality and you're like, nobody changed any of their behavior, and we just spend money on this tool for no reason, right?

And so how do you avoid that scenario? If you're able to again continually see what, what is actually happening, what are the workflows, you can track in real time, did we make the change that we said we were gonna do, and then did it yield the results that we thought it would or not? And then you can dynamically adjust, based on the results you're seeing.

That's incredibly... That's really hard to do manually. That's quite trivially to, trivial to do, once, once you start talking about using agents to do it. But I think that becomes really important in closing the loop. 

Richie Cotton: Absolutely. There's a lot of similarities there I think with with dashboards.

Like historically you have all these data teams create dashboards, and then do people actually look at them or not? And it's unclear whether you're gonna get a, an actual benefit from creating these things. 

Jennifer Smith: There's a lot of like great intentions in creating tools and agents on all of these things for other people, but then do They actually get used and did, did the, did they result in what you had hoped for?

Richie Cotton: Absolutely, yeah. So it does seem important to have that kind of retrospective on was this actually a good idea or not? So I guess in general, the hard parts can be like if you're doing something to save time or say, do people tell the truth about how much time they spend on particular tasks, or is this something I guess you're gonna automatically surface?

Jennifer Smith: People should not i- in the same way that like, creating documentation of that person sitting there and being like, "Okay, I'm gonna document what I'm gonna do." People should not be sitting there saying, "Oh, I spent three minutes doing this thing, like I should catalog it." That's a waste of people t- of people's time.

It's a cognitive overhead. That's something that should just be happening behind the scenes, right? And it should be looking across workflows, right? So this workflow more broadly done across these sets of people, like this is how much time it roughly takes right now and it's being done this way, and like we think it could look like this instead, right?

And then you can make a decision about whether you drive that. But you should not have individual people who are sitting there all the time thinking about "Oh gosh, how can I make this little thing better?" Pe- people are doing that naturally, right? I so my background when I was in consulting, I would go into these big operation centers, and we would just have a mandate like, make it more efficient, right?

And so our trick was we would just find the best person in that op center, and I would just go and ask them like, "Why are you better than everybody else?" And they'd pull out these thick binders with laminated pages of step-by-step guides. This was 20 years ago, they're digitals. The same thing today, okay?

They're just like online. There's no more lamination. And they would say, "I was trained on this. I don't do it that way, like I found a bunch of better ways," right? And so we would basically just create a new binder and we'd charge a million dollars a month for that. And it was totally crazy.

But the idea was pretty straightforward, right? Which is you can get everybody to do what your best person does, you raise the performance of everybody on the team. And so I think people always wanna be better at what they're doing at work. And they're constantly trying to find better ways to get things done, whether they even realize consciously that's what they're doing or not, right?

People are always gonna try to find the path of least resistance, like, how do I do this thing the fastest and the most accurately? I don't think it works if you just go and interview everybody and say what are you doing better?" They're just gonna look. It's, it's really hard for people to think that way.

But if you look across a set of people doing a set of processes and you can say, "Hey, we're doing this process 20 different ways, and like one of these ways actually looks really good. We should probably do it that way." Or there's probably some modifi- there's probably a, a different way entirely, that we could then standardize everybody off of.

But I don't think we should be putting that burden on individual people, like th- this is something that you can now do programmatically at scale. That's how we used to have to do it 'cause we didn't have another choice. 

Richie Cotton: No, yeah. I can certainly see how like filling out a time sheet for, oh, I spent three minutes doing this and seven minutes doing that, then that's gonna be incredibly tedious.

Jennifer Smith: We did a work with a b- a bunch of companies pre-AI, and I remember one company we went in and- They had their employees track their time in seven-minute increments for, like a period of a few weeks because they were trying to build this baseline understanding. And, the employees were actually pretty good sports about it because I think they were, like, pretty frustrated about how they had to spend their time in a lot of instances.

They're like, "We're doing a lot of things that we don't consider to be productive. They're just the proverbial BS, that we kinda have to get through. And so you're telling me that by tracking this, you can actually find the stuff. Okay, maybe I'll begrudgingly go along with it," right? But it was a really painful you can imagine.

Richie Cotton: I'm sure, yeah. I just spent seven minutes filling out a time sheet about what I'm doing. It seems like this whole area of a process analytics, process optimization is pretty fascinating. For people who are interested in a career in this area, like what do you need to know about it? Like what, what skills do you need?

Jennifer Smith: I think this should be embedded in the way that everyone works and everyone thinks, and I think this is part of what it means to be working in an AI native company now. I, again, as I said I don't think we need to ask everybody to suddenly learn how to code agents and all of these things that, you sometimes hear influencers talking about.

But I do think the best, the people best positioned, kinda career-wise in this new world are gonna be those who have a high degree of agency. And what I mean by that is people who think about what's the goal that I'm trying to drive here, right? If you think about what, in an organizational

I think about this a lot in just the way that I've designed our company and our kind of org structure is I think of them as locuses of accountability. Actually who's the person who's responsible for this thing going right? Or figuring out the best way to solve this problem. And so I think the...

and now what has changed? What has changed is people have way more powerful tools. They just have incredible leverage to be able to solve those problems, and so there's just far less resource and intelligence constraint in being able to do that, right? But I still think you're gonna see people as the locus of accountability, and so who's best in that kind of environment is people who are very high agency, who say "Okay, my job is to solve this big, broad, amorphous problem.

I'm gonna figure out the best way to do that, right? And I'm gonna leverage all the tools at my disposal, which may be headcount, which may be agents, which may be a mix of those two things together to make that happen, and I'm gonna be really thoughtful about the goal we're, the problem we're trying to solve, what the goal is, what the metrics are to measure it in, and then the best way, the best path for us to get there, and then how do we continually get better at that."

And so I think it's more of like a mindset and an ownership mentality and a critical thinking that comes with it versus a "I'm the best at figuring out how to prompt in Cursor." That's just that's just a means to an end. 

Richie Cotton: Okay. There's a certain irony there. It's to be the best human employee, you need more agency.

You gotta pretend you're an agent. 

Jennifer Smith: So I think it's deeply ironic that we... There's a lot of ironies. One of them is that we call agents, because I would argue the one thing they don't have is agency. I'm remodeling my house right now, okay? And I'm using agents to help me do it, but my agent can tell me...

I had to decide that it was time to remodel my house, right? Only I can decide that's, something that I want that's important right now. And then I decide what's the style that I want, what's the goal, what am I trying to achieve with this, why is this important to me, what are the constraints, what's the budget, what...

all of these kinds of things. I decide that. And then within that I can go to an agent and say, "Okay, help me figure out the best way to configure the bathroom," right? And it can give me a bunch of options, and then I still am still making the decisions ultimately of what it is that I want.

I, the thing I bring to that is agency. And increasingly, agents are gonna be able to take over more and more of those kinds of i- creative, generative tasks even over time. But I don't think they'll ever the replace the like, why are we doing this? Why is this important? Why now? What does success look like for me?

That still has to come from people. And so the best employees, I think, are gonna be the ones who exhibit those high degree of agencies who think of themselves as "I am the person defining why we are doing this, and for my part of the business or whatever I oversee, like why this is important, why we're choosing the things we do, why we choose not to do these other things, and how we're gonna get it done."

And then agents just become, again, incredible leverage on on those ambitions. 

Richie Cotton: So really these are like management skills, leadership skills of okay, deciding this is what we want to do 'cause this is how it's gonna improve the business, and then yet the the, the AI agent is then doing the work for you.

Jennifer Smith: You don't have the people management aspect of it that comes with management, right? So it's maybe management in the broader sense, if you think of a manager as being responsible for the output or the outcomes of a particular team. And if you define it that way, then yes, I think everyone should have a manager mentality, right?

E- even if it's just themselves. They're like, "I'm a team of one right now. I'm gonna manage the output of myself." But, pretty soon, we're starting to see this already, even if you don't have other additional headcount, you've got agents and other kind of resources at your disposal, and so it's how do I marshal all of these resources together to achieve whatever, my goal or objective is?

And how do I continually think about how we could be better at that? 

Richie Cotton: Yeah. I'm hoping it stays that agents don't start giving you people management troubles as well. Once your agents start calling in sick or complaining about work, then yeah, we got problems. All right. Do you have any final advice on how you go about documenting and improving all your processes?

Jennifer Smith: I think you need to have a really cl- again, it's, to me, it's all about harnessing specialized intelligence, and so what is it that makes our company special? What do we uniquely know how to do? What are the judgments, values, decisions that we make? Capturing all of that, harnessing all of that is a gargantuan task if you try to brute force your way through it.

You won't be successful. You can get slivers of it. That's what companies have been doing for decades. But to drive the kind of transformation at the speed that I think is going to be required, 'cause I think the, the pace of change is only accelerating I think maybe ironically, you have to use AI to help you do that.

And my advice to anyone is, use it to get a really clear understanding of what do we know how to do? What is special to us? And then use that to help define, how do we get better at it? And then as you are making changes, use it to then get better and to learn from what you are doing.

As humans do more work, as agents do more work, it becomes more input into the system, and so your system is now getting smarter and better over time a- as you're working. And I think that's the that's the big imperative for firms now. Everyone's talking about becoming an AI-native company, like whatever that means.

I think their big imperative is like, how do you harness this specialized intelligence so you own the things that are uniquely yours, and you compound in them over time. 

Richie Cotton: Yeah, I love that. Figure out what is unique about your own company, and then yeah lean into that. Nice. Okay finally, I always want more people to learn from, so whose work are you most excited about?

Jennifer Smith: I don't spend a ton of time on social media, but two folks that I like to follow on X. One's Aaron Levie from Box. Mostly when he writes about conversations with customers, it tends to match a lot of what I'm hearing in the field. And so it's just really great to hear, another perspective on, on what enterprise leaders are talking about right now, so I fi- I find that to be pretty straightforward.

And there's a woman named Jaya Gupta who's been writing a bunch of pieces recently that I've been following along these ideas of like, where are the moats in an AI world, and what's the role of generalized intelligence? She doesn't call it specialized intelligence, but like some of these similar ideas of what should the labs own and what should your company own, and what does that mean for what you should be doing right now?

Richie Cotton: Is it Jaya from Foundation Capital? Is this... Yeah. Yes yeah great stuff there. All right. Wonderful. Yeah. Thank you so much for your time, Jennifer. 

Jennifer Smith: Yeah. I really appreciate it. Thanks.