Earning AI Autonomy Through Governance With Wrike

[00:00:05] What happens when an AI agent can act, but nobody has decided just how much freedom it should have? Well, my guest today is the Chief Products Officer at a company called Wrike, where teams are bringing people and AI agents into the same working environment. And he will argue today that autonomy should be treated as a dial rather than an on and off switch.

[00:00:32] The question I find myself asking is, will that dial go up to 11? Well, agents earn greater responsibility as the work proves reliable. That is the thought process here, while permissions, review and human judgement keep higher risk decisions with people. So today we will discuss why prompts and token consumption are possibly vanity metrics,

[00:00:58] why impressive prototypes don't automatically create lasting business value, and why undocumented workflows leave AI short of context. So if your organisation is trying to move from AI experimentation to very real measurable results, then I'm hoping today's conversation will give you a useful place to start. And on that note, let me officially introduce you to my guest now.

[00:01:28] So thank you for joining me on the show today. Can you tell everyone listening a little about who you are and what you do? Yeah, hi Neil. It's great to be here. I'm a big fan of what you do. I'm a CPO at Wrike, and Wrike is the trusted work delivery platform. We built it for people and AI, and we have around 2 million users and a growing number of agents on the platform today.

[00:01:55] So what I think I can bring to this interview is just more of a practical view into AI adoption, because we do see it from the trenches. We work with customers, enterprise customers who are delivering their most important work on our platform. And in the process, they obviously go through the AI transformation. So they are looking for ways and guidance to enable AI within their workflows.

[00:02:22] And we are the vendor who is also trying to apply the same transformation internally as we do the work ourselves. So excited to talk to you about that, and many more. Thanks, Dave, I'm sure, are on top of mind feeling. And I'm excited to get you on here, because I think those real-life stories from the trenches are so important when we're discussing AI at work. There's so many people pulling it in the workplace, but not knowing where they are.

[00:02:50] Are they doing okay against their competitors? Are they going too fast, too slow? And I was reading a lot lately around the noise that surrounds Vibe coding. And I was reading that you'd argued that a Vibe-coded dashboard that is built in an hour didn't build a business, which really made me smile. And it's so true. So as AI makes it incredibly easy to create impressive demos and prototypes and, dare I say, dashboards,

[00:03:16] how do leaders listening distinguish between AI activity, AI theater, and technology that really delivers long-lasting business value? Yeah, that SaaS-pacopolis chatter has been, in my opinion, a bit overblown, to say the least. And I think they got the main assets wrong. It's not necessarily that AI can't help you build your dashboard, right? That's an easy part.

[00:03:46] And certainly you can use AI to build whole applications today. But if you compare that to like the real enterprise platforms and real enterprise applications that creates lasting value for our customers, those vendors, they do not ship the UI. They do not ship the features even. They ship the trust, right? The trust that their customers can get when they are running their own workflows.

[00:04:15] Business critical workflows on the selected platform, right? So and unfortunately, AI cannot substitute for that trust. In fact, it has to be governed. It has to be managed so that you can unlock the proper benefits of leveraging AI within your workflows.

[00:04:34] So I think it comes down to the fact that the models, which are getting increasingly more capable these days, they don't really understand how your organization does work. Right. So models can pretend that they know how this happens and they can certainly give you the very nice outputs of what you're asking them to do.

[00:05:00] But they cannot put the work that they're asked to do or that they do in the context of the overall enterprise workflow. And that's what I think really hinders the adoption of models and doesn't really allow enterprises or for that matter, any organization today to unlock the full benefit of AI. Yeah. So certainly they can build amazing things. They can build even the full applications.

[00:05:29] But again, do you trust the application that was invented by AI without having the appropriate context of how your work is done? That is the key question these days. Yeah, completely agree with you. And I think it was around six months ago, Yenson Wang, he hit all the headlines for the right or wrong reasons about measuring employee success on how many tokens they've burned.

[00:05:54] And I know you've talked about the end of token maxing where high AI usage can be mistaken for successful adoption. So what should companies actually be measuring if they want to understand whether AI is actually generating meaningful return on investment here? Yeah. And we see this stocking numbers ourselves, right?

[00:06:16] We did the survey recently and according to our research, only 80%, 34% of surveyed customers use AI consistently. We see that pretty much all the organizations like 90% plus have been finding AI, but only 34% have been using it consistently. And according to the other research, which McKinsey did recently, only 5% can actually say that they see high value coming from AI.

[00:06:45] So why is that disconnect? Why do we see this? Well, I think it starts with how you actually measure the value of AI. And I think you are very correctly pointing out to the fact that this kind of token maxing has been the wrong metric. I think actually usage in this case is a vanity metric.

[00:07:09] Brombs fired, tokens burned, a number of tools that you deploy to do AI. All of those are vanity metrics. And here's what people miss, in my opinion. You can't even see the ROI of what AI is doing inside the tool itself.

[00:07:29] Like when you are doing your work in a chat and you're asking me to do something, and obviously it is doing that for you and with decent quality even, you can't really tell how much it moves the needle. You can't really tell how much it translates to the organizational bottom line or even the value that you create in the process. You only see the output of work being created.

[00:07:54] And until you actually put the work that you can do with the help of AI in the context of a bigger enterprise workflow, right? Where there are multiple people involved, multiple tools being used, multiple outputs being created. And until you can actually see what business metric combined AI workload with people continuing to do the work and move, you will be blind.

[00:08:23] So what are the good methods? Good metrics are those that actually are found within the workflows themselves. So for example, if you are building something like an engineering product in a software industry, cycle time is a good metric. If you are managing client projects, you probably care about the customer satisfaction and your business margins. So those are the good metrics to prop.

[00:08:49] Now, it's not easy, obviously, to trace the impact of AI right from the top of this kind of value chain to the outposts that it produces for an individual on a daily basis. So you need something in between that could connect the work that you could do, an increasing amount of it with AI and the impact it actually brings to the bottom line of the organization.

[00:09:16] And I'm curious, from all the conversations that you're having and everything you're seeing and hearing inside workplaces that are experimenting with AI, why do so many promising AI initiatives stall when they meet real enterprise workflows? Is it because they're piling AI on top of old broken processes? And I've also got to ask, of course, what role do you see workflow redesign, change management and cross-team alignment?

[00:09:43] How do you see all these things turning experimentation into everyday adoption? You must get to hear so many different stories. I'm curious what you've seen. Absolutely. Yeah. So I hear a lot that you cannot really kind of like bolt on AI onto existing process, right? Like you have to pretty much re-architect your process so that it can write the benefits of AI, right? But I probably am going to share with you something like on the opposite side of this, which is,

[00:10:12] I do not necessarily believe you have to completely reinvent the whole process, right? I don't think that is the problem, to be honest. I think the problem is that even existing processes that you run within your organization are not properly documented, right? Are not even possible to describe, right? Because they are too unstructured. They are too silent, right?

[00:10:38] And when you introduce AI into the process, it is only as good as its understanding of how this process works, right? And we see our customers and we see obviously ourselves constantly re-explaining to AI that how the work should be done, right? And that is the core impediment to the productivity. You know, as long as you find yourself explaining or re-explaining how work should be done,

[00:11:07] that completely negates the time saved and the time that you receive by AI kind of doing the work for you. And I think basically, no matter what your process is, but if you cannot really describe it, if you cannot document it, if you cannot show the artifacts of the process, the historical artifacts of the process, the telemetry of work, as we say internally, AI will not be productive enough for you.

[00:11:35] You will always find a lot of the time being spent on reviewing its outputs, teaching it how it should be done, in explaining how things work here. So that's kind of one of you I wanted to share, but I also believe that change management is very important. Look at how humans organize their work today, no matter what, let's say, technology you use, right? No matter how perfect is your process, we are human beings, right?

[00:12:02] And we have to be led to actually perform a certain type of transformation. And we have to be given time to adopt to that new way of working. And the same is with AI. And when you infuse AI into your workflows, humans need to adopt, right? They need to understand what's the better and more efficient way of interacting with AI and where it is good versus where it is not good.

[00:12:29] And really perform that change management phase before AI becomes really productive. So those are the, in my opinion, two roadblocks today, which some of them are understood and appreciated, but nobody is ready to give enough time to just adopt to the new way of working. And some of them are, in my opinion, a combination of both the technology problem and the kind of work management problem at large within the organization.

[00:12:59] Yeah. And when it comes to AI in the workplace, you've encountered huge success here because you've helped launch autonomous AI agents to more than 2 million users. And as agents become part of the workforce, a quick question I think will be really beneficial to listeners here is how should they be defining their roles,

[00:13:23] their permissions, the boundaries and accountability, especially when humans still remain responsible for the outcome and accountable for the outcome there. And it is a big topic at the moment. We're hearing a lot around the topic of identity and AI agents, but any advice here? Yeah. In general, I think that organizations and employees need to treat AI agents as people, right? As workers in the kind of real workspace.

[00:13:53] And if you think about how we, for example, onboard employees, if we, if you think about how we get them up to speed and give them more autonomy over time and place responsibility on them and do the review of their work, all the same routines, in my opinion, should apply to how you work with agents. Let's take autonomy as an example. I don't think it's, it's, it's an on and off switch. I think it's a dial when it comes to AI.

[00:14:21] So you start by delegating things to AI that are helpful, but are not crucial. In other words, if AI agents makes a mistake, you can survive that, right? You can either redo the work it did or ignore the work it did, right? Or it's not, it's not a big deal in the first place if it made a mistake and you can just forget it. So, but then AI agents, just like humans, they need to earn more autonomy.

[00:14:51] Right? They need to show that they're capable of accomplishing more complex tasks, more complex work. And I think this is what the governance element really comes into play. The more governed environment you can create for the agents, the more autonomy those agents will receive over time. And conversely, the more complex work those agents will be, will be able to do.

[00:15:17] So let me give you a couple of examples, because we did launch our agents into our customer environments at the beginning of this year. And we were quite fascinated by how creative our customers can become with this technology and how many use cases they can actually find with that technology. And we see that initial adoption was for the use cases, such as status updates.

[00:15:42] Like you have a project and you want to have a nice summary of it without digesting all of its project plan. Another one is the pre-out of the requests that are incoming. It's a tedious work. Nobody wants to do this. And agents are perfect for that. So flagging risks as you execute initiatives, say, sorry. So, and we thought that this is where the adoption starts, right?

[00:16:09] Because those are high frequency, low risk tasks that you can delegate to agents. And indeed, we are seeing a great amount of adoption. But those types of use cases and with the first wave of agents, which we've launched. Now, where this is going to go next. In fact, we're building right now the next generation agents, which are going to be capable of handling more complex work.

[00:16:36] This is not just a simple output they need to create. This is goal-driven, right? They are given a goal and they need to find a way to deliver on that goal. And in the process of doing that, they can explore many different paths. And when they need feedback from humans, they will ask for it. And they are granted more autonomy in this case to do more complex work.

[00:17:04] But what enables this is this governance system, which we created around those agents. This is a system that manages their identities. This is a system that gives them only the right amount of access. This is the system that offers the so-called human in the low channel so that agents can actually receive the human feedback and they can have their work reviewed by the human beings.

[00:17:29] So that's what literally enables this higher autonomy, the governance. And that's what we're building within our platform. I don't think that governance is something that people should consider as an impediment, like something that you have to overcome. I think actually governance is what accelerates adoption of AI, right? Because it creates this checks and balances systems around what AI can do, should do, and is doing.

[00:17:59] And the sooner you find that system within your workloads, the sooner you will be able to delegate more work to AI. And ultimately, it'll be able to write more benefits from the work that it can actually do for you. Scale your business with agentic AI with limited risk. With Denodo's AI data layer, your agents are provided with real-time company data and guardrails for company protection.

[00:18:29] Create the business you always dreamed of with Denodo. And you can do that by simply visiting denodo.com to learn more. But now, back to my guest. That's such a great point there because not every AI use case should have the same level autonomy. So I think we've got to highlight that. And again, for people listening here, anything else you can add on how they can decide where an agent can act independently,

[00:18:55] independently, where human approval is required and where potential risk means AI should remain just an advisory tool. We've talked a little about trust and just give that a little bit more trust, a little bit more access, a little at a time. But anything else you can share there? Because I think it's a question a lot of people are going to be asking at the moment around what it should be allowed to do and what still requires a little trust and a bit of human intervention.

[00:19:22] Yeah, I think there are simple rule of thumb that you can share here, right? First is take a look at the work that you hate to do every day. So obviously, that's the work that is usually low value, that is usually high frequency, and that is usually something that you would like to completely automate and delegate. So this is where you would start to build that trust.

[00:19:46] And in our world, it can range from just like you and I are talking right now, and we are capturing notes from these meetings, right? So having AI summarize this, digest this, refine this, and to build into a nice structure is the type of work that can be delegated in the first place, right? And there's a number of examples of that nature. But then if you start to think more about it, like, okay, so I got that one handled. I found more capacity to do a higher value work.

[00:20:16] How can I use AI here, right? And then you think about using AI as your thought partner. Can AI actually help me vet my decisions? Can AI help me perform the research and fill in the blanks on the topic that I'm doing right now? So this is where AI needs to understand more context of what you're trying to do. This is where AI needs to be almost like your co-worker, right?

[00:20:42] And you need to really work with that AI within the context of the same task or goal that you are trying to achieve. And the higher up the stack of that work you go, the more impact is that you can find. And also the more, the higher is the cost of making mistakes. And I'm a true believer that at the end of the day, judgment has to be reserved for the human beings, right?

[00:21:08] And AI is a helpful tool, is a companion of the work that you should do, both as an individual worker and as part of a team. But the high-stakes decisions, those that were a number of, let's say, people depends on, or where the whole trajectory of the business depends on, they need to be reserved to the human beings. And that is why I'm really big on this kind of human to AI collaboration.

[00:21:38] So I do think it shares the same traits of how humans collaborate with each other. And that's why we are introducing AI into our platform in exactly the same ways as humans are already represented there. But at the same time, AI has to be treated in a special way. You always have to be ready to mitigate the negative impact of what AI can do, right?

[00:22:01] And have these checks and balances in place to recover from the mistake AI did, right? Or just prevent those negative ramifications in the first place. So that's how I think about it. And as different teams and indeed individuals adopt their very own AI tools and agents across large enterprises, there's also a risk of fragmentation and fragmentation at machine speed, you might argue as well.

[00:22:28] So what foundations around context, governance, security, and oversight are needed to just ensure that that environment is created where AI can operate reliably across the entire organization without creating some of the old problems around fragmentation and silos, etc? Great point. I think what we hear these days is that there's an explosion of different kinds of tools, especially now that you can trade your own tool with AI.

[00:22:58] And I think that the too many tools is not the root cause, it's the symptom, right? The root cause is that AI and people, they need to reference the same context, right? And the context is scattered around different places, different tools, different repositories. And solving that problem, which is aggregating context and making sure that everybody can access it from the same vantage point is a very, very tall task.

[00:23:25] Not only because you would have to somehow find a way to connect at all, but because you also need to kind of bring it under the common denominator, like structure it in the right way and surface that context in the right places of your workflow. So, and that's what really is the hardest problem to solve. And we solve that through the work intelligence graph. So work intelligence graph is what we have invested into for almost like 20 years.

[00:23:53] And it's not just the data structure. It's not just the way to aggregate and host the vast amounts of data. It also is the semantics layer that understands who is involved in this workflow, how the different stages of this workflow are structured, who is accountable, who is something that, who has access to what within this particular workflow. So all of that is being recorded as you do your work.

[00:24:23] And the more of your work you put through the workflow, the more that work is captured in the work intelligence graph. And it creates this kind of compound of effect, which we call the organizational memory. So basically the more iterations and the more cycles you go through the smarter and more productive every participant of the workflow becomes, including agents themselves.

[00:24:49] So we're not afraid of agents entering the workflow because all of that activity is going to be captured back into the work intelligence graph, which conversely will make save agents smarter over time because they are going to be learning from their own interactions with the humans and from their own outputs and the reviews that they're getting within that workflow. So context problem can be solved.

[00:25:16] But you really need to have the right structure to put that context into and really need to make sure that that context will be maintained in the way that creates this compounding advantage versus in the way that creates friction. So, and that's what we have been thinking about and thinking about and building for years.

[00:25:38] And I'm really excited to see that the moment is finally here where organizations can start to find it as a compounding benefit for themselves. Well, we've covered a lot today around Wrike, complex work delivered by humans and agents, AI in the workplace. So for anybody listening, where's the best place for them to find you or your team online and find out a little bit more about anything that we've covered today? Where would you like me to point people listening?

[00:26:07] I would start with our own kind of website, Wrike.com. So usually the latest and greatest pieces such as research announcements, new product is advertised on that website. I would also encourage people to tune into our annual conference, which is Wrike Columbrate. This is exactly where we're going to be talking about the effect of transformation and the challenges for transformation when people are looking for right ways to work alongside AI.

[00:26:37] And obviously, I'm pretty active on LinkedIn and all of the kind of thought leadership pieces that come from us. They're usually featured on LinkedIn. So please check it out, both from myself and from other Wrike leaders who are bringing this great product, great platform to the thousands and tens of thousands of customers these days.

[00:26:59] Wrike is the trusted platform that is helping scale what teams and AI agents can deliver with full control and clear accountability at every step. So I will be including a link to the website so people listening can find out more information. And I'll also include not just a link to your profile, but a great article that you wrote on LinkedIn or a post there about, I think it was the line, a vibe coded dashboard built in an hour, didn't build a business.

[00:27:29] I mentioned it earlier in the podcast and I think it's a great read too. So I'll put all that into the show notes over at Tech Talks Network. But a big thank you for taking the time and sitting down with me today and bringing all this to life today. Really appreciate it. Thank you, Neil. It's been a pleasure. Thanks so much. I think my guest's advice today gives leaders a very useful sequence for bringing AI agents into everyday work. Yes, begin with frequent tasks.

[00:27:58] And if there's an error that is easy to correct, that is great. And watch the results. Build the confidence. And then only increase autonomy or turn that dial slowly up when the system has earned it. If you run into problems, maybe just turn that dial down a little again. And this approach changes how we think about governance too. Permissions, human review, shared context and recovery controls.

[00:28:23] These are just a few of the mechanisms that lets an organisation delegate work in a responsible way. And the other lesson today I'm going to be taking away is around measurement. Tokens, prompts and tool counts. They all tell you that people are active. But cycle time, customer satisfaction, project margin and completed work. These are the real metrics that show you if AI is creating value. But over to you.

[00:28:49] Where would you turn the autonomy dial first in your organisation? Would you go straight up to 11? I would hope not. Or will you start nice and steady and just slowly go from one to two to three and get to where you need? Let me know. TechTalksNetwork.com Remember, I'm on the road a lot this year. I'm going to a lot of tech conferences. It'd be great to meet you in person. I don't just say this at the end of every episode. So I generally get a real kick out of meeting you all.

[00:29:18] That's why I do this. So go to TechTalksNetwork.com, go to events, have a scroll down and hopefully I will get to meet you then. But that is it for now. I'll be back again soon. But thanks for listening. Bye for now.

[00:29:46] Scale your business with agentic AI with limited risk. With Denodo's AI data layer, your agents are provided with real-time company data and guardrails for company protection. Create the business you always dreamed of with Denodo. And you can do that by simply visiting Denodo.com to learn more. But now, back to my guest.