On March 8, 2021, a committee at Credit Suisse set up to watch risky clients sat down for a 90-minute meeting about six of them. One was a hedge fund called Archegos, whose exposure had grown dramatically. This time the committee wrote down a plan, and next to it a name: the head of the business’s own risk unit was designated the “owner” of the action. He was given a target date of April 2021.
The committee had first taken up Archegos’s breaches the previous September. It decided then to come back to them “at a future meeting,” set no date for fixing anything or even for reporting back, and did not discuss the fund again for nearly six months.
Now there was a name and a date. Less than three weeks later, Archegos told the bank it could not meet its margin calls.
Who’s the DRI on that?
At Apple, Fortune’s Adam Lashinsky reported in 2011, “there is never any confusion as to who is responsible for what.” The company even had a name for that person: the DRI, or directly responsible individual. “Any effective meeting at Apple will have an action list,” a former employee told him. “Next to each action item will be the DRI.” When people at Apple wanted to find the right contact on a project, the question was simply: “Who’s the DRI on that?”
GitLab’s public handbook, which credits Apple with the term, describes the DRI as the person “ultimately accountable for the success or failure” of a project, with “final decision-making power.” The DRI is expected to consult the people affected and gather their input. They don’t have to win everyone over before deciding.
A DRI is not the person who does all the work. On a new GitLab feature, for example, the product manager is the DRI for what gets built first and the engineering manager for getting it delivered. The point of the name is narrower. If this goes wrong, it tells you where to start asking.
A name is not enough
Credit Suisse’s problem was not a shortage of people. The board’s investigators found that the leadership of the unit serving Archegos made things worse, “with neither Co-Head believing he ‘owned’” its US swaps business, and neither actively managing it or its risks. Two people can share a job if they split it clearly. Here nobody had.
And when the committee finally wrote down a name, the date next to it came after the danger. The bank lost about $5.5 billion; the case has the details. TSB’s IT migration, three years earlier, ran into a smaller version of the same thing: when reviewers asked who carried the risk of key testing decisions, they found it unclear.
In practice, a DRI works best with three things attached:
- A name for each piece. Co-heads or “the team” can own work, but only once it is clear who covers what.
- A date that matches the risk. “At a future meeting” is not a date, and neither is one that falls after the trouble arrives.
- Authority within a clear scope. The power to decide what is theirs to decide, and a direct line to escalate what isn’t. An owner who must win every argument before moving is just a reporter.
Naming one person doesn’t let everyone else off. The team, and the organization above it, still answer for their parts.
The chores nobody owned
Imagine five friends sharing a flat. The rubbish goes out “whenever someone notices it’s full.” Everyone notices. Everyone assumes someone else will do it, or did it yesterday. By Friday the kitchen smells.
Then they pin a rota to the fridge: one name per week. The person whose week it is doesn’t have to carry every bag. They just have to make sure it happens. The rota doesn’t guarantee a clean kitchen. But when the bin overflows, everyone knows whom to ask, and the conversation can move on to what got in the way that week.
A person behind every agent
AI agents make it easy to forget this. An agent can run a task from start to finish, and it is tempting to say the agent owns it. But an agent cannot answer for a decision the way a person can, and when something goes wrong, the question still lands on people.
For high-risk systems, the EU’s AI Act points the same way. Organizations deploying them “shall assign human oversight to natural persons who have the necessary competence, training and authority, as well as the necessary support.” The law doesn’t ask for a single owner, and it doesn’t hand the whole outcome to one person. It does insist that whoever oversees the system can understand it and has the power to step in. A DRI is one way to organize that.
The same habit works for everyday agent work. For each task an agent runs, write down which person owns the result, when they will check it, and what they are allowed to stop. If that line is blank, it will be hard to say afterwards whose job it was.