Giving AI agents names and a spot on the org chart makes people feel less accountable for the work, and it squeezes the agent into a role it doesn't need.
Dave Sharrock and Peter Maddison dig into new research from the September/October 2026 issue of Harvard Business Review on what happens when organizations add named AI agents to the org chart. When people use AI as a tool, they own the result and check it carefully. When the same work comes from a personified agent, accountability quietly shifts to the software, and nobody is really holding it. Peter connects this to the idea of accountability sinks, the layers in an organization where responsibility disappears. They also talk about review fatigue, the growing pile of AI-generated documents, slides, and code that someone still has to read, understand, and sign off on. Onboarding an agent like a new hire still makes sense. Leaving it on the org chart with a name is where the trouble starts.
This week's takeaways:
- How people work alongside capable AI agents is still poorly understood, and the old problems of ownership, accountability, and decision making do not go away just because an agent is involved.
- Review fatigue is real and getting worse, because cheap generation still means expensive human review for every artifact someone has to approve.
- Naming an agent into a human role limits it to the shape of that role, while treating it as a tool in the workflow lets it work across boundaries in ways a person cannot.
Listen to the full episode at definitelymaybeagile.com
Subscribe so you never miss an episode.
Have a question or topic you'd like us to cover? Reach out at feedback@definitelymaybeagile.com
New episodes released every Thursday to challenge your thinking and inspire action.
Listen and subscribe:
Welcome and Topic Setup [0:04]
Peter Maddison [0:04]: Welcome to Definitely Maybe Agile, the podcast where Peter Maddison and Dave Sharrock discuss the complexities of adopting new ways of working at scale. Hello, Dave. How are you today?
Dave Sharrock [0:15]: Excellent, Peter. Good to catch up. How have you been doing?
Peter Maddison [0:18]: I'm doing very well. I'm a little bit tired today, but I'm going to try and bring my A game with lots of energy, and I'm sure you're going to bring a really energetic topic for us to talk about.
Dave Sharrock [0:29]: Of course we're going to bring an energetic topic. So I've just been catching up with the Harvard Business Review, the September/October 2026 issue. Like every magazine, every place you read now, Harvard Business Review has tons around AI. They've got a big section on treating AI basically as a new hire: how to bring AI agents into your organization in a way that dovetails as easily as possible with how you stand up teams in your organization.
When AI Joins the Org Chart [1:10]
Dave Sharrock [1:10]: I didn't really want to talk about that part, but they also have an article citing research on the impact of having agentic AI in your org chart structure. Named AI agents, if you like. What impact does that have on the rest of us working around those agents in the org chart itself?
Peter Maddison [1:38]: Yeah, and we were talking just before this about what to talk about today, and we were going through this topic a little. I do find it interesting. One of the pieces you brought up from this article was that when you introduce agents as named entities in the org chart, with specifically defined roles that are bounded by what the human role they're supposedly doing would be, you start to take the agency away. And I love that. Agentic agency, agency for agents, or some combination of those words, because it's just fun. The English language is lovely like that.
Dave Sharrock [2:18]: It's helping us with that one, yeah. I found it interesting because there's a lot of talk right now about agents and agentic AI. We think of agents as independent actors in a system, and those agents are typically people, you and I, as employees or contractors in an organization. The interesting takeaway from the research, from my perspective, is how it impacts our feelings about our role and how accountable we feel for the work that gets done by these personified or non-personified agents. There's a very real difference between using AI through a command line interface as a tool to help me do things, whatever that might be, and handing over a problem to a named agent in my org chart and expecting that agent to give me back a deliverable that I might then hand off to another team.
Peter Maddison [3:25]: Yeah, and I think this is purely a human psychology piece. We feel threatened. As we have talked about many times before, people can feel threatened because this is attacking their identity and how they see themselves. Am I still valuable? Am I still needed in this role? And that gets amplified when it's an independent, named agent doing the work. Even though the other side would argue that by giving it a nice fluffy name... actually, that's a great name. Fluffy.
Trust, Identity, and Leadership Clarity [4:03]
Dave Sharrock [4:03]: So if we call the agent Fluffy... What really came out was that interesting bit around trust and undermining my confidence or my positional identity in an organization. A lot of that stems from unclear or obscure communication from leadership about the purpose of these AI agents and tools. We talk about this so many times when we look at transformations. Clarity of purpose and direction from executive leadership about why a change is happening is really critical. It helps stem the uncertainty and insecurity that naturally come with questions like: is my role under attack, and how is it going to change? That just gets exacerbated when you can see these named AI agents in the org chart. They sow seeds of concern about my own position in the organization and where my role might be going.
Peter Maddison [5:08]: Yeah, and I think it would take us in the wrong direction if we started talking about what drives that. There are a lot of potentially false promises and hype driving a lot of those behaviors and that uncertainty. So there isn't necessarily a clear line back to why we are doing this in the first place.
Dave Sharrock [5:31]: Well, how do we put it? There is a difficulty when executive leadership is strategically trying to move in a particular direction. For example, if they're going to lead the organization toward reducing headcount, or reducing training or opportunities for employees, there's always a challenge in how you talk about what are effectively negative strategic directions.
Peter Maddison [6:02]: Yeah, exactly. But if we leave the connotations to one side and think about how something gets introduced to the organization, introducing an agent as a named member of the workforce matters. One of the other pieces that caught my eye in that article, and I agree with it, is that this also limits how the technology gets used. If I create an agent that looks like a person, it gets the role boundaries of that person. Whereas if I take a step back, the agent isn't a person. It can take a very different shape within the system. It could span multiple areas or boundaries and operate differently depending on how you need the system to behave. So instead of the round peg in the round hole, you're looking at a very malleable thing that could take all sorts of different shapes.
Dave Sharrock [7:09]: Yeah. When you describe it like that, it opens up a really great opportunity to say this agent isn't replacing an individual in a role. We're not moving an agent in and moving a person to one side. It's more that this agent can have a breadth of understanding and skills across the organization that we could never get from individuals, so it can solve real problems that wouldn't be easy to solve any other way. That starts building a bridge to, okay, we can see why that agent is doing it. And I think part of the article says: don't give it a name and put it in the org chart.
Peter Maddison [7:53]: Yeah. Because then you're just scaring people.
Dave Sharrock [7:56]: Yeah, exactly. But that's not the only thing the article talks about. It touches on two other things, and we talk about both of these all the time. One is the reduction in review quality, and the other, which we'll touch on separately, is accountability.
Review Fatigue From Content Floods [8:13]
Dave Sharrock [8:13]: If I start with the reduction in review quality, we have both experienced it, and we talk about it whenever we get on a call together. We're overwhelmed by the volume of material we're now expected to read through and approve. Whether it's a prompt, context, skills, or output from these various AI agents, there's a lot of content to review. And reviewing it with the right level of cognitive attention is just overwhelming. It's a real shift in the roles we're often playing.
Peter Maddison [8:52]: Yeah, and it does differ significantly. We're not just talking about code review, which has been talked about ad nauseam. It's the volume of documentation, the volume of slides, the volume of content you're now looking through and asking: is this the right messaging? The right direction? The right format? Is it going to communicate the message the way I want it to? And it's coming at you much faster than before. Generation being cheap doesn't really help you, and that's the point. It's still expensive for a human on the other end to absorb and understand it. Yes, you can have agents absorb it, refine it, and get to a better version, but at the end there's still an artifact that somebody has to say yes or no to.
Dave Sharrock [9:51]: And be accountable for. It's not just reading it and saying, okay, the English reads well. It's reading it and asking: is the logic behind these conclusions correct? Is it in line with everything we know? That's hard if we haven't been actively involved in analyzing and really understanding those recommendations. It's like me reviewing your recommendation to a client. I can look through it, but to actually understand every recommendation, I have to dig deeper to make sure I agree with the direction you're recommending, not just that it reads well.
Peter Maddison [10:35]: Yeah, exactly. It can appear to read well very easily if you're just skim reading, so you need to read it in more detail. That takes time and bandwidth. You need enough time to sit down, read it, absorb it, and understand whether it aligns with what you want to say. Sometimes you skim it and go, oh, that's terrible.
Dave Sharrock [11:02]: Invariably that's what I start bumping into on LinkedIn. You can spot it very quickly when certain phrases come up. We're getting very good at identifying things that are mostly AI-created and haven't been polished.
Peter Maddison: For sure.
Dave Sharrock: Now, that brings us to the final point I wanted to touch on, which is accountability.
Accountability Shifts to the Agent [11:22]
Dave Sharrock [11:22]: One of the really interesting things the research did was compare two situations. If I use AI embedded in a tool, or through a command line interface where I'm just prompting and getting output, then as an employee I feel accountable for the outcome. I review it with a clear eye for accountability before I pass it on to the next person. But if there's a named agent in the org chart, this personification, I shift the accountability to that agent. I no longer feel responsible or accountable for the outcome.
Peter Maddison [12:14]: Yeah.
Dave Sharrock [12:15]: So either I escalate it and say I'm not sure this is the result we should be looking for, or I just let it go by because I don't feel accountable for the outcome. That's a problem, because it's a piece of software. It isn't an individual who can be accountable for the outcome.
Peter Maddison [12:36]: And it was hard enough to get people to be accountable for things anyway. Half the time they ended up being accountable for completely the wrong thing. So this doesn't really help. There's the concept of accountability sinks, which is the disconnection from true accountability produced by the extra layers we put into an organization. It's the person at the flight desk who can't rebook your flight because they don't have the power to do it. There's no point getting angry at them, because they're not the person you should be angry at. And I think this takes that and magnifies it, by putting accountability sinks throughout your entire organization and giving them names.
Dave Sharrock [13:27]: Well, yeah. The really strong takeaway was: don't personify your agent. I don't think the article states it quite that boldly, but when I look at it, this is tied to how we understand roles and how we interact with people. We're not purely rational beings. We have emotions, and we feel threatened by agents with people's names sitting where we used to have colleagues. And then there's the question of how responsible we are and whether we'll be held accountable, and we tend to push that away from ourselves. These are natural responses. So the key takeaway is just don't do the personification.
Peter Maddison [14:17]: Yeah. And as you say, that's quite contrary to a lot of what's out there. Does the article talk about what they recommend doing instead?
Use AI as Tools, Not Coworkers [14:25]
Dave Sharrock [14:25]: Well, the research is based on using AI as tooling versus using AI as personified agents with names. With tooling, we still take accountability. It's like using a calculator and wanting to double-check our working. We use a tool that helps us do complex things, we know we need to pay attention to the results, and there's still a connection there. But as soon as we personify it, the difference is substantial in how much we let things slide and lose the attentiveness we would otherwise have.
Peter Maddison [15:12]: Yeah, I think it logically makes sense. An agent is effectively a prompt being fed into a model and running at that level. The idea that the personification exists as anything other than a tool doesn't hold up. It can go and take action for you, but really it's the ability to say, "@ name of agent, do this," or kick off the job, or pick up this ticket and run with it. That kind of thing.
Dave Sharrock [15:51]: The interesting thing about this issue of Harvard Business Review is that there's a whole bunch of content saying treat AI like employees in terms of how you set them up. And I think there's a lot of value in what they discuss. This article doesn't say don't do any of that. It just says don't name it at the end. If you onboard an agent, treating it like a new hire so it understands the constraints, the governance requirements, and the rules your organization operates by completely makes sense. And the way they talk about briefing agents like really smart interns also makes total sense. The only issue is don't leave them as named interns or agents at the end. Keep them in that tool perspective, and you get the benefits of setting these agents up correctly without the impact on accountability, trust, and identity we discussed.
Three Takeaways and How to Respond [16:52]
Peter Maddison [16:52]: Yeah, that makes a lot of sense. So with that, what three points shall we leave our audience with today?
Dave Sharrock [16:59]: I'm going to pick a couple and leave you one. Two things jump to mind. Number one is that the way we as human beings interact with these very capable agents is not fully understood at all. There's so much more to learn, and I'm sure that research will continue, whether it's about judgment, trust, identity, accountability, or decision making. We know from working with employees and transformations how difficult it is to get ownership, accountability, and decisions made in the right places. None of those problems disappear. They still need to be recognized and addressed from a human perspective. The other one, because we glossed over it, is review fatigue. Every one of us struggles with it, and it's a difficult one because it's not going away.
Peter Maddison [17:56]: No, it's going to get worse before it ever gets better, if it does.
Dave Sharrock [17:59]: Yes.
Peter Maddison [18:00]: So I'll pick up the third one. I like the point that naming agents into roles in an organization means you start shaping them into what that role is, and that may not be the best shape. If you think of an agent instead as a tool within the workflow of the system, it can take a very different shape than you originally imagined. Thinking about how you can use it within the system you're building is a broader and more solid way of thinking about these agents, and it will be a lot more valuable in the long run. So with all of that, I'd like to thank you as always, Dave. And to all our listeners, don't forget to hit subscribe, tell all your friends, and send us messages at feedback@definitelymaybeagile.com. Looking forward to next time.
Dave Sharrock [18:52]: See you next time. Thanks again, Peter.
Peter Maddison [18:54]: You've been listening to Definitely Maybe Agile, the podcast where your hosts Peter Maddison and Dave Sharrock focus on the art and science of digital, agile, and DevOps at scale.