Deploy free

Bot vs human commits: watching AI vs human commits shift week to week

Bot vs Human Commits tracks AI vs human commits at the unit that actually ships: non-merge commits, split into those carrying an AI-assist trailer — a Co-Authored-By line from a tool like Claude Code — and those that don't, plotted as two weekly lines across GitHub, GitLab, and Azure DevOps. Commits, not reviews, because a commit is work that actually landed.

deckgauge · Board · Intelligence
Bot (trailer-tagged) commitsHuman-only commits

Bot commits climbing, human commits roughly flat.

What it measures, and where the data comes from

Every non-merge commit in the window gets checked for a recognized AI-assist trailer. Merge commits are excluded on purpose — they're the git mechanics of combining branches, not authored work, and including them would dilute the ratio with commits nobody actually wrote.

  • GitHub — commits on connected repos, trailer-checked, merge commits excluded.
  • GitLab — same check, across every connected project.
  • Azure DevOps — same check, from the repos you've opted into code intel for.

Both lines are counts per week, plus a bot-share % in the summary tile, so you can read either the raw volume trend or the relative split.

The trailer is a floor, not a ceilingThis counts commits with a machine-readable AI-assist trailer the tool actually stamped. It says nothing about commits where someone used an assistant that doesn't leave one, pasted generated code by hand, or had the trailer stripped by a squash. A low bot share can mean genuinely low adoption, or it can mean invisible adoption. Read this as "the share of commits we can prove are AI-assisted," never as "the share of code that is AI-written" — and don't present it to a VP as an exact figure.

How to read it

Look at the two lines relative to each other, not the bot line in isolation. A rising bot line next to a flat human line means AI-assisted work is additive — the team is doing the same human-authored volume plus a growing AI-tagged slice. A rising bot line next to a falling human line is a different story: AI-tagged commits may be substituting for human-authored ones, which is worth understanding before you decide whether that's good news.

What it tells you over time

A steady divergence — bot climbing, human flat or dipping — that lines up with a specific week is almost always traceable to a tooling rollout or a policy nudge ("start using X"). A slow, gradual climb with no clear trigger point usually means organic, peer-driven adoption rather than a mandate. Both are real signal; they just call for different follow-up questions.

Example situations

1. Bot commits climbing, human commits pinned dead flat — a capacity question

deckgauge · Bot vs Human Commits
Bot commits — steady climbHuman commits — pinned flat

Bot commits nearly tripling over ten weeks; the human line hasn't moved by more than a commit or two.

What you're seeing: this isn't the usual "AI adoption growing on top of steady human output" chart. The human line isn't just roughly flat, it's barely varying at all — genuinely pinned, week after week — while the bot line does all of the growing. When total output (bot plus human) is rising and none of that growth is coming from the human side, the story is about capacity, not enthusiasm.

How to react: don't let this alone justify freezing a req or skipping a backfill. It can mean two very different things: a team already at full capacity where AI is genuinely absorbing new work nobody had bandwidth for — a good story — or a team that quietly lost headcount (attrition, a freeze, a reorg) where the bot line is masking a gap rather than adding on top of a healthy baseline. Also sanity-check the trailer itself before either conclusion: if a tool update or a config change stopped stamping Co-Authored-By reliably around the same window, some of what reads as "flat human" commits may actually be untagged AI-assisted work, which would understate both lines' true split.

Managerial playPull headcount for the same window before drawing any capacity conclusion from this chart alone. Flat human commits next to stable or growing headcount is a productivity story worth defending in a budget conversation. Flat human commits next to headcount that quietly shrank is a gap being papered over — worth surfacing before someone assumes the team can keep shrinking and the bot line will keep covering for it.

2. A high bot share that isn't showing up anywhere else

deckgauge · Bot vs Human Commits
Bot commits — high, holdingHuman commits — unchanged

Bot share above 40% for two months — but Bug Rate and Review Quality Index haven't budged.

What you're seeing: bot-tagged commits are sitting above 40% of the total — a genuinely high, headline-worthy share by most teams' standards. On its own, that number reads like a win. But pulled up next to Bug Rate and Review Quality Index for the same window, neither one moved: bugs are flat, review comment rate and instant-merge rate haven't budged. A high bot share is proof of usage, not proof of outcome.

How to react: don't let a big bot-share number stand in for "AI is making us better" in a report — that's a different claim, and this widget doesn't measure it. Volume of AI-tagged commits tells you the tool is being used a lot; it says nothing about whether that usage is producing cleaner code, fewer regressions, or faster review. Those need their own widgets, checked directly, not inferred from this one.

Managerial playBefore citing this number as evidence of impact, pair it with Bug Rate and Review Quality Index for the identical window and look for movement in either direction. If both are flat or worse, treat that as a real question, not a footnote — ask the team what the AI-assisted commits are actually being used for (boilerplate and scaffolding move this number without touching quality; logic-heavy changes would). A high bot share and an unchanged quality picture together are a prompt to look closer, not a reason to declare victory.

Frequently asked

How do you track AI versus human commits?
Non-merge commits are split by whether they carry an AI-assist trailer, such as a Co-Authored-By line from a tool like Claude Code, and plotted as two weekly lines across GitHub, GitLab and Azure DevOps.
Why measure commits rather than pull requests?
A commit is work that actually landed. Reviews and PR-level markers describe process; commits describe output, which makes them the more honest unit for tracking how much code is genuinely AI-assisted.
Does a low bot share mean the team is not using AI?
Not necessarily. It commonly means untagged adoption: the tooling in use is not adding a trailer. Confirm how your assistants are configured before reading a low bot line as low usage.

Related widgets

  • AI Adoption — the same trailer detection joined with the PR-level marker, side by side by period.
  • AI-Assisted PR Percentage — the PR-level version of this same signal, count-weighted across the window.
  • Commits per Developer — check this alongside a rising bot share to see whether it's changing individual output, not just the mix.
  • Back to the widget reference.

Last updated