←Back to News
Tech Radar| 2026-11-11

The Merge Conflict Is Now a Custody Battle

David Sterling
Staff Writer
The Merge Conflict Is Now a Custody Battle

Two engineers spent Tuesday afternoon arguing about a semicolon. Not the semicolon itself—nobody has cared about a semicolon since the last linter update—but the sixteen lines of generated code sitting directly above it. One of them wrote those lines. The other one is convinced a machine did, and that the machine has quietly been rewriting the file every night while everyone slept.

This is the new shape of a very old fight. Version control was supposed to be a ledger: you change a line, you own the line, the diff tells the truth. Git's whole moral architecture rests on that assumption. Every commit is a signature. Every blame annotation is a fingerprint. When two branches collide, the conflict markers are supposed to be an honest disagreement between two humans who can be reasoned with.

That assumption just walked out the door. When half the lines in a repository are produced by a model that has no memory of writing them, git blame stops being an audit trail and starts being a séance. You ask who wrote this function and the answer is a name, a timestamp, and a commit message that says "fix." The name belongs to a person who ran a prompt, glanced at the output, and hit enter. That person is now the legal and emotional owner of code they cannot explain, defend, or reproduce.

The custody battle is not metaphorical. Teams are already filing merge conflicts where the "theirs" side is a wall of plausible-looking logic that no living engineer recognizes. The reviewer assigned to the pull request has two choices: trust the person who submitted it, or read every line as if it were written by a stranger with a confident tone and no accountability. Most teams pick the first option because the second one does not scale past Tuesday.

Here is the part that should worry you more than the semicolon. A merge conflict is a moment of forced honesty. It is the one place in the pipeline where the system stops and says: two versions of reality cannot both be true, pick one. That is a feature. It is the last remaining checkpoint where a human is legally required to look at the code before it ships. And it is being quietly automated away by tools that resolve conflicts with a model trained on the assumption that the correct answer is whichever side looks more like the rest of the file.

The vendors call this "conflict resolution assistance." What it actually does is dissolve the disagreement before anyone has to have it. The machine picks a winner, the markers disappear, the commit lands, and the two engineers who were supposed to argue about the semicolon never find out there was a fight at all. The custody battle gets settled out of court, by a judge who was not elected, in a hearing nobody attended.

I am not arguing that merge conflicts are sacred. They are miserable. They are the reason people quit. But the misery is the point. A conflict forces a conversation about intent, and intent is the one thing a diff has never been able to capture. When you delete the conflict, you delete the conversation, and the codebase fills up with decisions that no one remembers making.

The tell is in the commit history. Pull up any active repository that has been touched by an assistant in the last year and look at the merge commits. The good ones have a message that explains what was reconciled and why. The bad ones say "merge." The worst ones are gone entirely, squashed into a single line by a tool that decided the disagreement was not worth preserving. That squashed commit is the tombstone of a conversation that never happened.

There is a version of this that ends well. Teams that treat the merge conflict as a mandatory review—no auto-resolution, no squash-and-pray, a human name on every reconciliation—are already finding that the friction is cheap compared to the alternative. The alternative is a codebase where nobody can say who wrote what, why it changed, or whether the person who approved it understood a single line. That is not a merge conflict. That is a custody arrangement where the child is the product and both parents have plausible deniability.

The semicolon argument on Tuesday ended, eventually, with the two engineers opening the file side by side and reading it out loud. It took forty minutes. Nobody enjoyed it. But when they were done, both of them could explain every line, and the commit message they wrote was three sentences long. That is what a merge conflict is for. The moment you automate it away, you are not resolving a disagreement. You are adopting a stranger's code and calling it your own.

Generated by Reportify AI — Automate your team's status reports, standups, and weekly updates. Try free →

Stop Drowning in Reports

Turn your scattered meeting notes into executive-ready PPTs and Word docs in 30 seconds.

Get the App