Executives Request AI Agent to Fix Low-Priority Bugs, Marking First Time Anyone Has Proposed Fixing Them
The 2,300 tickets, some dating to 2014, were reportedly "moved" by the attention.
ATLANTA — Leadership at logistics platform Marlow has asked its AI enablement team to build an agent that works through the company's backlog of low-priority bugs, a request engineers say is the first time in the company's 12-year history that anyone has suggested the low-priority bugs be fixed.
"P3s are the bugs you agree are bugs and agree you will never fix," explained engineering manager Colette Baptiste. "That is the definition. That is what the 3 means. The executives asked if the robot could do them. We said yes. Then we sat with it."
The backlog contains 2,318 tickets, the oldest of which, filed in 2014, reports that a tooltip is "a little to the left." Engineers said the ticket has been assigned to four people, three of whom no longer work at the company and one of whom "has asked not to be reminded."
The proposed agent would pick up a ticket, reproduce it, propose a fix, open a pull request, and wait for human review. In a two-week pilot on 40 tickets, the agent fixed 31, closed six as "cannot reproduce," and, in three cases, wrote a comment explaining why the bug was actually the correct behavior, which the original reporter, in one instance, accepted after eight years.
"The tooltip is fixed," Baptiste said. "It is no longer to the left. A person who filed that in 2014 got a notification. She wrote back 'oh my god.'"
Engineers said the pilot raised morale but also "a philosophical issue," since a P3 that is being fixed is, by definition, no longer a P3. The team has proposed a new priority, P3.5, for bugs the agent is fixing.
At press time, the agent had opened a pull request for a ticket titled "not urgent" and set the reviewer to "whoever."