A conversation with David Mallon of Deloitte on why the tech-first approach to AI is backfiring, what 88% of leaders are getting wrong about orchestration, and the design choice most organizations don’t realize they’re making.
The Moment You Can’t Sit Out
David Mallon has spent 15 years helping track the currents that shape how people work. As a research leader at Deloitte, he co-authors the firm’s annual Global Human Capital Trends report, one of the most comprehensive looks at the human side of organizational life produced anywhere. This year’s edition frames 2026 as a moment where tensions have become tipping points, and the framing is deliberate.
“We’ve all had that moment where we stood in a river or in the ocean and felt this pressure pushing you somewhere else,” he told me. “That’s what this report is about.” For the past several years, that pressure has been building around questions that leaders have struggled to answer cleanly. What is a job in an AI-enabled world? Who counts as part of the workforce? Where does work actually happen? Most leaders, faced with those questions, have been doing what leaders tend to do when the answers aren’t clear: waiting for more information.
That option, David says, is no longer available. “Trying not to make a decision is a decision,” he said, “and it’s usually the one with the least likely chance of success.” The compression between when a choice has to be made and when its consequences land is getting shorter. The fence that leaders have been sitting on is disappearing beneath them.
Why the Tech-First Approach Is Backfiring
The report contains a finding that should give pause to anyone who has approached AI in the same way they approached the last major technology rollout. Organizations taking a tech-focused approach to AI are 1.6 times more likely to miss their return on investment than those leading with a human-centered strategy. And yet 59% of organizations are still defaulting to the tech-first path.
David’s explanation for why is generous but direct: “They’re misunderstanding the moment.” Most organizations looked at AI and assumed it worked like previous technology waves, cloud, ERP, the usual succession of enterprise software. Buy licenses, do change management, expect good things. “That’s not AI at all,” he said. The more accurate comparison, he argues, is to electricity or steam power, general purpose technologies that took generations to fully reshape how work was organized. “We’re probably not going to spend a generation figuring this out, but it’s the right frame.”
The practical implication is that buying tools and nudging people to use them is not a strategy. A European telecom company in the report illustrates this precisely. When they added an AI tool to their customer service operation without redesigning the work around it, productivity improved by 5%. When they rebuilt the entire workflow with AI at the center, productivity jumped to 30%. The tool was the same, but the design made the difference.
For smaller organizations without transformation-level budgets, David’s advice is to think in capabilities rather than transformations. “Where can you work with leaders at every level to use your imagination about how those workflows can change?” The goal isn’t to solve one pilot problem but rather to think about how horizontal capabilities, the ones that run across functions and get used for many different things, can evolve in ways that bring the workforce along rather than leaving them to figure it out alone.
The Work Slop Problem
One of the more uncomfortable findings in the report is what happens when AI gets distributed through an organization without any real thinking about skill development. The term David uses, borrowed from broader market conversation, is work slop: the passable but shallow output that emerges when less experienced workers use AI to produce results they don’t fully understand.
The research suggests AI widens the gap rather than closing it. Experienced workers using AI get sharper. Less experienced workers produce work that looks adequate but isn’t. And the longer-term consequence of that dynamic is a workforce that hasn’t actually developed the judgment to know the difference.
This connects to what David calls the experience gap, a trend the report tracked last year. Organizations have been flattening their structures and reducing entry-level roles for years, which means fewer of the foothold positions where people used to build foundational experience. AI has accelerated that trend. “What used to be the requirements of a manager are now the requirements of a first-time hire,” he said. The result is that organizations are quietly hollowing out the pipeline of experience they’ll need to draw from in five years.
Some organizations are responding by deliberately building friction back into the system. Moments of practice without AI tools. Apprenticeships and internships making a quiet comeback. Entry-level roles being reinstated not because they’re efficient but because they’re necessary. “If everybody has a diamond-shaped organization,” David said, “you’re not going to hire your way out of this. You’re not going to go find somebody else’s experienced worker.”
The Cultural Debt Nobody Budgeted For
Perhaps the most striking number in the report is this: 80% of leaders and workers are concerned that their teammates are using AI to look more productive than they actually are. That level of suspicion, left unaddressed, does predictable things to organizational culture over time. Trust erodes, collaboration becomes transactional, and people start optimizing for the appearance of performance rather than its substance.
David’s diagnosis is that most organizations simply skipped a step. They deployed the tools without ever having an explicit conversation about what the tools change. “What’s hard work? What’s cheating? What’s the difference between cheating and augmentation?” These questions sound almost philosophical until you realize that without answers, your workforce will develop its own answers, and those answers may not match what you intended.
The promotion question is particularly thorny. Do you advance the person who worked 30 hours and mastered the technology, or the person who worked 60 hours without it? “These seem like basic questions,” David said, “but they deserve to be talked about out loud in organizations.” Because if they aren’t, the existing reward and recognition systems, built for a pre-AI world, will keep running quietly in the background, reinforcing behaviors that may no longer align with what the organization actually values.
His recommendation isn’t to pause the AI rollout. He suggests that leaders run the cultural conversation in parallel with it. As workflows get redesigned, ask what the human’s role is in each piece of work, and what they’re actually bringing. “It’s not activity,” he said. “And increasingly it’s not creating knowledge, because the tools do that really fast. What else is it? Curiosity. Asking the right questions. Connecting dots the systems can’t connect. If that’s what they’re bringing, how do we know they’re doing that well?”
Decision Rights in an Agentic World
The report includes a quote from a leader that David found worth highlighting: “The moment AI enters the workflow, the real question isn’t ‘what does the model say?’ It’s who gets to disagree with it and how fast.” This is a useful reframe for a conversation that tends to get stuck on capability rather than accountability.
Most organizations, David observes, have never been particularly rigorous about decision rights outside of major capital allocations or board-level governance. For everything else, decision-making authority tends to be assumed, or worked out through relationships, or left to whoever is most persistent in a given meeting. That informality was always a little costly. But with AI agents in the picture, it becomes a real liability.
“Human agency is where accountability comes from,” he said. When something goes well or badly, if it matters to know who’s responsible, you have to engineer that line of sight into the system before the decision is made, not after. With agentic AI especially, where a chain of automated actions can produce outcomes that no single person explicitly chose, that line of sight can be genuinely difficult to construct. Organizations that haven’t thought carefully about decision rights are going to find that problem waiting for them.
Have Organizational Functions Outlived Their Function?
One of the more provocative questions the report poses is whether traditional organizational functions like HR, finance, IT, or legal still serve the purposes for which they were designed. David is careful to say the question is mostly rhetorical. Functions aren’t disappearing, but the pressure on them to clarify their purpose is real and growing.
Part of what’s driving it is that the boundaries between what finance knows, what HR knows, and what legal knows are becoming harder to justify when the underlying data is all sitting in the same systems. “Finance data and HR data and legal data,” he said. “Why do we box things the way we box them?” The case for domain expertise in each function remains strong. But the case for those domains operating as insulated silos is getting harder to make. AI doesn’t respect the walls organizations have built around their data, and neither do the business problems organizations are trying to solve.
What David is seeing in the most forward-looking organizations is the emergence of what he describes as a trio: one expert in how humans behave and what motivates them, one expert in AI technology, and one expert in whatever domain the work actually involves. “You’re going to need all three working together versus any one just doing it sequentially.” Where that trio lives in the org chart is still an open question. Who owns work, in the abstract sense of the word, remains genuinely unsettled.
The Design Choice Leaders Don’t Know They’re Making
As we closed our conversation, David’s last point is one that I think deserves more attention than it typically gets in discussions about AI and the future of work. He referenced an economist at Stanford named Eric Reynolds, who observed that the role of humans in organizations is a design choice. Not an inevitability, not the output of market forces, but a choice.
“We have a lot more agency over all of this than I think we sometimes realize,” David said. “There’s this tendency to think we’re all just on an amusement park ride and along for the journey.” The report’s final chapter, titled “Decisions That Echo,” makes the point that board-level choices about how AI gets deployed ripple outward through communities, families, and civic trust in ways that extend well beyond the quarterly earnings call.
His ask of leaders is straightforward. Think about what you want the future to look like, because you are already making that choice, either through deliberate action or by default. “We’re all choosing which science fiction movie we want our future to be,” he said. The organizations most likely to get this right are the ones that stop treating AI as a cost reduction exercise and start treating it as a question of what kind of value, and what kind of work, they want to exist in the world.
That shift, from efficiency to design, from automation to intention, is the one the data says most organizations haven’t made yet. It’s also, David would argue, the one that’s actually worth making.
Brandon Laws is a workplace culture and leadership enthusiast, host of the Transform Your Workplace podcast, and VP of Marketing and Product at Xenium HR.