The Human Cost of Disconnected AI Agents

As a leader tasked with scaling AI, you've been promised that these new tools will revolutionize your organization, but the reality on the ground feels different. You're told AI should streamline your operations, yet your employees are operating as human APIs, forced to serve as the manual integration layer between disconnected tools that don't share context. Today, we're unpacking exactly how to solve this friction and move past a fragmented reality of siloed workflows and agents. We're handing the mic to an expert in meeting users right in their flow of work, Leila Al Mounir, Vice President of Connected Experiences at Workday. Here's Leila on how to design a future where your technology finally speaks the same language. Hi, I'm Leila Al Mounir, Vice President of Connected Experiences at Workday. I oversee how products can best meet users in their flow of work But much of my career has been spent at the edge of the next big thing. Many years ago, I was designing product experiences during the massive transition to mobile-first retail. Back then, I was leading the strategy for mobile, and my team noticed something fascinating. Shoppers would obsessively browse a retailer's mobile app, add items to their cart, and then drop off. They wouldn't buy. At first, we panicked. We investigated if it was a mobile issue, a problem with the UI, or a funnel conversion issue in checkout. But it wasn't. We realized these shoppers were using the app to essentially pre-shop. They were browsing digitally but still choosing to go into the store. They wanted to touch the fabric and check the fit before they made the decision. It's hard to believe now when most of our shopping is done completely online these days. But that was my first lesson in interoperability. We had an opportunity to design an experience that would meet their needs by bridging the gap between the digital in the physical world. We capitalized on the problem and turned it into an opportunity by offering capabilities like reserving an item and trying on in store, or buying it online and picking it up in store, or even helping them locate the item in the store. Ultimately, we improved conversion rates in the ways that met the needs of the user and drove business results. Today, we're facing a similar challenge, but the landscape has changed. We aren't bridging mobile apps to stores, We're bridging a traditional digital experience to one that integrates and elevates agents, users interacting with agents, and even agents interacting with other agents. Today, I'm gonna teach you how to capitalize on the new opportunity in front of us and move past the fragmented reality of siloed workflows and agents. You'll learn the value of interoperability and how to unlock productivity and value from your organizational data. To understand where we're headed, we first have to understand the friction we're feeling today. Until now, specialized applications have been our primary unit of value, and we've spent the last two decades capitalizing on that value. The software was designed to do one thing exceptionally well, whether managing your inbox, your ledger, or your supply chain. Interoperability between these applications isn't a new concept. We see it every day when your calendar automatically syncs a flight from your email, or when your Slack notifications pop up in your browser. These tools learned to speak to each other to save us time. But while these systems learned to share data, they didn't learn to share context. They gave us a head start, but they still left the actual execution up to us. For years, this was the status quo, a web of specialized applications that required us, the humans, to act as a primary bridge. We were the ones who opened the browser, logged in, found the data, and moved it. But this kind of orchestration puts the onus on the user to figure out which technology to invoke to answer a specific question or complete a specific task. It's like having 10 remote controls for one TV. Imagine trying to watch a movie, but one remote handles powering it on and off, another handles the play button, and a third handles the volume. You spend more time managing the remote than actually watching the movie. And today, this is going to be even more critical. We're entering a new phase where we aren't just deploying more applications, we're deploying AI agents. And while specialized agents is powerful within its own domain, true enterprise value requires more than siloed expertise. It requires agents that can discover, communicate, and coordinate across systems. The utility of these agents is multiplied when they meet us in their existing flow of work, whether that's in a desktop or a mobile application or in the collaboration tools employees are already using. Let's take a look at a scenario every leader knows by heart. Meet Sarah, a manager preparing for a talent conversation with her direct report, Marcus. Sarah believes that Marcus is ready for promotion and she needs the data to back it up. Without interoperability, Sarah has to go on a digital scavenger hunt to build a compelling justification of his impact. Consider how a manager preps. They're often having to dig through Word documents for one-on-one notes, HR platforms for goals, and messaging platforms for feedback. This manual bridging of tools is the friction. By the time Sarah has gathered all of that siloed data, she might be too cognitively drained to put it in writing or even think about how to justify a promotion. This is likely a factor in why, according to Forbes, only 14% of performance reviews are found effective by the direct report. This is where we must shift the narrative to interoperability as a value driver. True value is unlocked when agents can share context and data from multiple sources to deliver combined insights. Imagine Sarah's experience in a connected enterprise. Instead of a scavenger hunt, she asks a single agent, "Summarize Marcus's performance over the last 6 months." The agent doesn't just look at the one database. It coordinates with the documentation platform, the goal tracking system, and the feedback tool to execute a cross-functional workflow. This brings us to the importance of an open ecosystem. You have to decide: do you require a hyper-personalized walled garden experience where you own every part of the orchestration? Or do you prioritize for an interoperable and open ecosystem? The walled garden approach gives you a perception of total control, but it risks becoming tedious and siloed. The open approach leverages interoperability protocols and APIs to allow agents from different domains or even third-party applications to work in harmony. This isn't just a build versus buy decision. It's a decision about how much friction you are willing to let your users tolerate. We need to stop asking users to navigate apps and agents separately and design an architecture that brings data and action directly to the user right in their natural flow of work. To truly solve for this friction, we have to move towards an open and interoperable platform that still honors your security and privacy requirements. To achieve this and build a truly connected enterprise, we need a roadmap that prioritizes architecture over individual features. We have to recognize that we are no longer just managing a human workforce. We are managing a hybrid ecosystem of both people and agents. Just as we have a system for onboarding people, we need a system of record that defines the agents, what data they can touch, and how they interact with people. Without this system, we aren't building a connected enterprise. We're just managing a collection of expensive and disconnected agents. So what are the steps you can take to begin building the foundation of interoperability? The first is to take inventory of every AI deployment in your stack. Identify the data silos they touch and map the friction points where users are forced to manually bridge the gap between agents. Think of these as swivel chair moments where your employees are forced to act as the integration layer. Let's return to our manager, Sarah, to see this in action. As she prepares materials for Marcus's review, she might use one agent that summarizes peer feedback, another that tracks project velocity, and a third that manages skills and certification. Every time Sarah has to locate an insight from one agent, switching screens to check another, and then manually bridge the gap just to decide on a course of action, for example, a promotion for Marcus, she's acting as a human API. When you audit your footprint, you aren't just looking for a list of tools. You're looking for where your most valuable resources, your people, are being used as the connective tissue for your data. If the intent captured in Sarah's performance summary doesn't flow automatically into the action of a promotion for Marcus, that is a high-friction handoff that needs to be solved at the architectural level. And these are key moments of opportunities you can capitalize on and solve for immediate value gains. Once you've conducted your audit and identified these areas of opportunity, you're ready to move on to step 2. The second step is how you build for open interoperability. Every enterprise is approaching this challenge differently. You might be a massive organization building your own custom orchestration layer, integrating with a third-party AI-ready platform, or leveraging AI solutions you already use in your stack. Regardless of which path you choose, the key to solving the scavenger hunt is making sure your ecosystem is open, connected, and interoperable. Instead of building massive data infrastructure from scratch, you do this by adopting emerging industry standards. Like Model Context Protocol, otherwise known as MCP, or agent-to-agent communication. These are opening up the possibilities of connecting data and agents from completely different sources. These protocols handle the handshake, the verification process that allows Agent A to securely coordinate to a multi-step process with Agent B without a human middleman. Without them, your agents will never be more than isolated experts. With them, they share a common language. But a protocol alone is not enough. You also need a transparent collaboration layer where every agent decision, every action they take, and every piece of context is used, logged, governed, and auditable. At Workday, we call this Agent System of Record. Whether you build it yourself or demand it from your AI vendors, prioritizing open protocols and a centralized system of record turns interoperability from a plumbing problem into a strategy for intelligence. The final step is to decide where to build and where to connect. This is about resource allocation. If you have massive engineering resources and workflow that is core to your competitive advantage, you might build your own orchestration layer to own every pixel of the user experience. But for high-velocity standard workflows, you can consider leveraging open protocols and APIs to allow agents from different vendors to work in harmony. By building this foundation, you ensure that as the technology evolves, your enterprise is ready to speak its language. Now, let's talk about confidence and trust. You can connect every system in your company, but if you don't build a layer of human confidence on top of it, your users simply won't use it. That is because true interoperability presents a unique psychological challenge. When systems talk to each other autonomously, the actual work becomes invisible to the user. In an enterprise setting, that can often breed anxiety about trust and the risks of hallucinations—how accurate the data really is. Think about the complexity of travel miles as an example of this gap. Imagine you have elite status with one of the airlines, but you're booking a flight on a partner carrier. The second you hit confirm, you're going to immediately check if the systems talk to each other and if your status carried over. That anxiety exists because the handshake between the two systems is invisible to you. You are waiting for a signal that the interoperability actually worked, and perhaps it's failed you in the past, so you always have a tendency to always want to validate if it didn't go wrong again. In the enterprise, the stakes are even higher than lost travel miles. If a payroll agent updates a highly sensitive pay change in an HR platform, the user needs to know without a shadow of doubt that the information landed exactly where it was supposed to. If they don't see that confirmation, they'll revert back to their old manual habits or be required to validate the data, which adds time because they don't trust the ghosts in the machine. To solve for this, here are 3 steps for creating an infrastructure of trust. First, cite the source. If an agent pulls data from a third-party document, it must tell the user exactly where it found it. Don't just give an answer, give the receipt. Second, preserve the brand. Our research shows that without a branded stamp of authority like a logo, users feel a deep sense of distrust about whether the data is officially stored. Seeing that familiar branding tells a user, this is an official update. Third and finally, implement a verification loop. When agents perform an action across systems, it must provide a proof of work notification. It's the digital equivalent of that airline app finally showing you your miles as confirmed. When these three layers work in harmony, we move from crossing our fingers to seeing a clear narrative of impact. Let's return to our manager, Sarah, preparing for the performance evaluation to see how this framework comes to life. At first glance of the data, Sarah can easily see the citations that indicate where the data points were pulled from, along with the timestamp to be sure that the assessment is on this fiscal year's performance. She sees the branding, giving her assurance that this data is speaking to the right tools, or in this case, HR application. And finally, when she's ready to submit the promotion, she sees the multi-system verification process in action. She receives a confirmation message that tells her exactly what was updated, where it was updated, and what the next step is. This infrastructure doesn't replace Sarah, it equips her. The data and process was no longer a distraction. That is the ultimate promise of the connected enterprise— technology that works so seamlessly, it finally allows us to be human again. Sarah is now free to do only what a human can do— lead. Trust is the lubricant of interoperability. When you provide clarity and verification, you move the user from a state of anxiety to a state of agency. Now, think back to those shoppers that I mentioned at the beginning of this lesson. The friction has evolved. It's no longer the gap between the digital and the physical world. It's an internal friction in the digital and human system. We see it with our manager, Sarah. Drowning in a sea of browser tabs and intelligent agents that don't talk to each other. When we architect for interoperability, we are finally solving that friction for Sarah. We are moving from a world where you have to navigate the software to a world where the software navigates the enterprise for you. As you move forward from this masterclass, step back and reflect on where your organization actually sits on this journey. Ask yourself these 3 questions. First, have you moved beyond basic automation? Are you truly deploying agents that truly simplify the task at hand, like generating deep content from rich data? Or are you just adding another tool to the pile? Second, do you have the mechanism to manage them? As your agent footprint grows, do you have the infrastructure to oversee their actions, or are they becoming a new shadow IT problem? And third, the most critical question: Have you allowed your agents to interoperate so your users don't have to? Can your agents interact with one another to remove the burden of disconnected enterprise data? Or are you still asking your humans to be the integration layer? The connected enterprise isn't a destination you reach overnight. It's the standard you build towards, one handshake at a time. The choice you make today about your architecture and trust will define your organization for the next decade. Let's build a future where our technology finally speaks the same language. Thank you for joining me. That was Leila Al-Mounir on the critical importance of architecting for agent interoperability. Isolated AI agents might supercharge specific tasks, but they can't connect a fragmented enterprise. And as a leader, your challenge isn't just deploying more technology, but intentionally building the open ecosystem and infrastructure of trust that stops your people from acting as human APIs and keeps them operating in their natural flow of work. Thank you for listening to the Future of Work podcast. Be sure to subscribe so you don't miss great episodes like this one.