When AI Does the Work, Who Is Accountable? – Somehow Manage

Skip to content

Everyday I find myself grappling with the rapid adoption of AI and what it means for human ownership over the work we deliver and accountability for that work. As I think about how we’ve been approaching this at Monarch so far, I wanted to share my thoughts and hear from others.

My father’s family is from the Egyptian town of Ismailia, which sits right in the middle of the Suez Canal. Growing up, we would watch as the massive ships navigated the narrow waterway while small, more agile boats sped up alongside them. As the vessels matched each other’s speed, a local Egyptian pilot would climb aboard and assist the ship in navigating the canal. Every ship navigating the Suez Canal would take on one of these local pilots.

If the ships already had their own captains and crew, why did they need local pilots? While the captain and his crew continue to own the ship’s ultimate goal — getting the ship, crew, and cargo to its destination — the local guides have two goals:

To use their superior knowledge of specific technical and local issues to help the ship navigate the Canal.

To maintain, through care and coordination, the health and function of the overall Canal, for many ships across multiple years.

At Monarch, we’ve always leveraged a system of Project DRIs and Domain Owners that in many ways mimics what I witnessed growing up. In practice, this looks like:

Project DRIs that focus on execution of a particular project. Their goal is a successful outcome. They and their team are the ship’s captain and crew. We’ve always aimed to give Project DRIs as much autonomy as possible, and with AI, those Project DRIs can now execute a lot faster and more flexibly. Imagine our ship’s crew above, but suddenly, that ship’s speed has quadrupled and it has gained the ability to traverse not just water, but land and air.

Domain Owners that have a deep understanding of a particular “domain” just like the local guides. Domain boundaries aren’t strict, and can be drawn in different shapes and sizes. For example, a domain could be our product’s overall visual design, our budgeting code, or a part of our technical infrastructure.

Like the local pilots, domain owners maintain a deep understanding of their area coupled with a strong sense of ownership and accountability for it. They help others navigate their domain while ensuring that domain is healthy over the long run. It’s the ownership and accountability that ultimately results in good work.

Can AI be a domain owner?

AI does not have the capacity to be a domain owner. At least not yet. To do that, it would need the right knowledge — both domain expertise and context — and the right accountability loop.

As LLMs get better, they will have better knowledge and domain expertise. Their “jagged edges” will smooth themselves out. And as we get better at handling the “mismanaged genius,” we’ll find the best ways to give them the context they need. It may take some time and effort; because, unlike AI, humans have the ability to recall things that are important to them while ignoring things that are not, and to fill in missing or bad context with expertise or common sense.

The one thing LLMs won’t likely have is the essential degree of ownership and accountability. If you’ve been part of a high-performing team you know that so much of the work succeeds because of human trust, accountability, and initiative.

I want to be clear that accountability for high-performing teams is not of the "who’s neck do I wring if something bad happens" variety. It’s the confidence you have in work that has someone you trust’s name on it. Someone who is competent and conscientious and has some sort of (material and/or reputational) skin-in-the-game. For example:

“I trust that my colleague understands their work and will do it well, and if they don’t, given feedback, they’ll correct it, because their name is on it.”

“I trust that if I ask my colleague to do something that doesn’t make sense, they’ll push back in correlation with their level of conviction (rather than pander to me).”

“I trust that my colleagues will take initiative and come up with creative solutions to the problems we have.”

If we introduce AI thoughtlessly and haphazardly, we risk undermining the very parts of human ownership that actually result in great work and differentiate great work from work that is just okay. No matter how much knowledge or context AI has, it may never replace a well-informed, well-intentioned human with a sense of ownership.

Avoiding the slop trap

Throwing copious amounts of information (code, text, design, etc.) at someone, without the ability to judge the quality of that information, is bad citizenship. It just so happens that LLMs make it easier to produce copious amounts of information that looks good at first blush, but puts the onus on the reviewer. This is where ownership and accountability matters most.

We still...