drumline

Does AI make your team ship better?

The measured story of the AI era: commits are surging, but far less of the work reaches a release (NBER w35275). Drumline reads a GitHub repo's history and shows where it lands on the write-to-ship curve, and how it compares to its peers.

Generate a report for my GitHub repo

Public repo, read-only, no setup. A human replies.

From other GitHub repos

The most telling reads, names withheld.

✅ Lock this in
Adoption up, reliability held

One repo's AI-assisted commits tripled in a week (4% to 14%) while its CI failure rate fell by half. More agent work, and a steadier pipeline, not a shakier one.

💡 Do more of this
AI writes, humans still gate

One repo now writes most of its code with an agent, but still reviews and runs CI by hand: no agent reviewed a pull request or triggered a run all week. The write side raced ahead; the gate stayed human.

✍️ Make the AI measurable
One signal survives

A commit trailer is the one AI signal that survives a squash and a rebase, and most agent commits do not carry one. Add Assisted-by: or Co-Authored-By: and the impact stops being a guess.

👍 Hold this
7× the volume, still shipping

One repo's pipeline volume jumped sevenfold in a week and its CI stayed mostly green. When agents flood the pipeline, this is what holding up looks like.

💪 Keep this fixed
The worst gate, 77% to 9%

One repo's flakiest gate went from failing 77% of its runs to 9% in three weeks, and the wasted minutes fell with it. Whatever the team changed worked; the job now is holding it there.

💡 Two signals, one story
More hands, more merges

One repo's author base widened to 44 in a week, and they did not just push commits: 53 changes merged, 43 of them onto main. Community growth that ships, not noise.

What the benchmark reads

Your repo on the write-to-ship curve.

Actionable insights

What your repo says about shipping with agents.

Agent impactWhether agents are making delivery better: failure and re-run rates as their share of the work climbs, side by side.
By agentWhich agent is worth keeping? Copilot, Cursor, Devin and your own bots, ranked by what they ship and cost.
What to fix nextEvery read ends on one move, not a wall of charts: the workflow trending the wrong way, its multi-week arc, and the single fix worth making before next week, like quarantining the flaky suite sitting on your critical path.
Pricing

Start free. Pay when you'd miss it.

Open Source
$0
free forever
The foundation:
  • Public repositories
  • Write-to-ship benchmark
  • Weekly email report
Start with Open Source
Team
$99 /mo
billed monthly
On top of Open Source:
  • Private repositories
  • Group multiple repositories
  • Unlimited team members
  • Deliver to Slack
  • 3-month history
  • Monthly trends
Start with Team
Business
$299 /mo
billed monthly
On top of Team:
  • Real cross-vendor cost
  • Cost centers & budgets
  • Cheapest-runner recommendations
  • 12-month history
  • Quarterly trends
  • Priority support
  • API access & export
Start with Business
Custom
Talk
tailored to you
On top of Business:
  • SSO / SAML, SCIM, RBAC
  • Audit logs
  • Self-hosted
  • Your data region
  • Custom retention
  • Custom runner rates
  • SLA, invoicing, procurement
Talk to us
FAQ

Questions, answered.

What does Drumline actually tell me?+
Where your repo sits on the write-to-ship curve: how the surge in agent commits and PRs converts into merges and releases, and whether reliability holds as the agent share climbs. All from public GitHub history, read-only. The Business plan adds your real cross-vendor CI cost, private to you.
What is the write-to-ship curve?+
The measured gap between writing and shipping. Across 100,000+ developers, stronger agents raised commits by up to 180%, but only 30% of that surge reached actual releases (NBER w35275): the review and merge stages are where the gains leak. Drumline shows how much of your repo's writing actually ships, and how that moves week over week.
How do you know what's AI-authored?+
From the signals that survive: commit trailers (Assisted-by:, Co-Authored-By:), known agent and bot identities, and the actor that triggers each run, bucketed into people, bots, and AI agents. It's a floor, not truth: inline completions leave no trace, so the real share is at least what we show. Every report says so.
How do I get the report, and who else sees it?+
It arrives weekly by email, with a web link that expires after 7 days. It's yours: we don't publish it, and no cost number ever appears publicly or gets compared across repos. Paid plans keep the accumulating archive so the trend builds instead of expiring.
Do you need write access to my repos?+
No. Drumline is read-only. It reads workflow and usage metadata and never modifies your code, settings, or runs.
Does it work with GitLab or other CI?+
GitHub Actions today. That covers GitHub-hosted runners and third-party runners like Depot, Blacksmith, and Namespace; self-hosted runners get time attribution with your own $/minute. Other forges are on the list, but we would rather do one deeply than three shallowly.
How is the cost calculated?+
Cost is a Business-plan feature. We reconstruct it from job metadata (duration and runner type) priced at each vendor's current rates, then reconcile against your monthly usage export. On public repos it's a modeled equivalent (no bill exists); on private repos it's your real billed spend. Always private to you, never public or compared across repos.
Can I cancel anytime?+
Yes. No contracts, no demo call to start, cancel in two clicks. Annual plans are billed at a 15% discount.