Blog
The Nudges Are Only as Good as the Data Under Them
We shipped a stack of nudges that check in on an agent's work automatically. Most of them depend on knowing which tests cover which files, data we hadn't tested against real code until we pointed Lien at nine real open-source projects and read what it actually said.
The Blind Spot Every Reviewer Has, and the Bug That Proved It
A popular open-source database toolkit added six new column types. Four companion packages were never updated to handle them. Our automated reviewer read all the right files and still missed it, most of the time.
Does Warning a Coding Agent Actually Change What It Writes?
We warned an AI coding agent, right before it touched a function, that the function was already hard to follow. Then we tested whether that actually changed what it wrote, or whether the agent just ignored it.
Why We Built Lien
A tool that gives AI coding assistants a real, structural understanding of your codebase: what depends on what, how complicated each piece of code is, and what tests actually cover it. Runs entirely on your own machine.