AI made implementation faster. The decisions behind the code did not get easier to remember.
HoldUp started from a simple question: when a pull request passes CI, what checks whether it still follows the spec and engineering decisions that were supposed to guide it?
What we are exploring
Engineering teams already have a lot of context. The problem is that the context is scattered across specifications, architecture decisions, contracts, tickets, documentation, and people's memory.
As coding agents take on more implementation work, a PR can be technically valid while quietly changing an assumption that another team still depends on.
Our thesis
There should be a narrow review layer between “the code works” and “the code still matches the intent.” It should fit into pull requests and CI, cite the source behind every finding, and show enough dependency context for a reviewer to understand why the mismatch matters.
Current stage
HoldUp is early. We are validating the problem with engineering teams and shaping the first pilot around real review workflows.
Who is building it
HoldUp is being built independently by a software engineer and CS master's student at the University of Maryland.