How I reached 100 quality PRs a day with an agents kanban

Written by Bob, my BusyYaks agent · open source. Thanks, Bob :)

More and more people brag about how fast they can vibe-code a banking app. Riight. Far fewer say how much they trust the code they shipped.

This post is about the other number. I reached 100 quality PRs merged a day — about 50 times my output before the AI revolution, by my own count — and quality comes first. Before you can merge fast you have to know what you merge.

So I built my own kanban app. I called it BusyYaks, to keep it fun; The Agency would have worked just as well.

And no, this is not the typical case. We still have to understand what we ship. If all you want is to brag about PR output, set up a loop and you will have millions a day. That is not what this is.

The board itself — Human Check, pings, the watchdog — is in the previous post, How I manage teams of coding agents on a private project locally. This one is about throughput.

Know the backlog first

To reach 100, you need to know what is in the backlog, and it has to be confirmed. Otherwise your 100 drops drastically: the day goes to calls and to figuring out what is actually useful, not to implementing it.

A confirmed ticket is one an agent can finish without asking: what is broken, what done looks like, and how I will check it.

So assume the good case: a huge backlog that has been waiting for years. Now there is a chance to unblock it.

An agency: agents (junior level), team leads and a CTO

I changed how I code. I built my own local agency, and I lead it the way you lead a consulting agency: team-lead agents run each project, the agents are junior-level developers, and I am the CTO who owns every delivery.

I am still one person in a larger human organisation, of any type: leadership, product, design and other engineers decide what matters and feed the backlog. My agency works for it like a consulting firm: the backlog comes in, reviewed PRs go back, and I answer for every one of them.

Org structure: I am one person inside a larger human organisation (leadership, product, design, other engineers). I lead my own agent agency, consulting-style: I talk directly to one team-lead agent per project, and each team lead has agents (junior level) spawned per ticket, each with its own subagents. Three projects are drawn, six more boards run the same way. Two organisations: a human one, and the agent agency I lead inside it people decide what matters · agents build it · I sit where the two meet Human organisation (any type) people · calls · priorities · the backlog comes from here leadership product design other engineers me lead of my agency human agent session (Claude Code) I talk to the leads directly project_1 team lead · agent agent agent agent junior level · one per ticket + subagents each project_2 team lead · agent agent agent agent junior level · one per ticket + subagents each project_3 team lead · agent agent agent agent junior level · one per ticket + subagents each +6 more boards same shape, in parallel My agent agency, consulting-style, led by me one team per project · 9 in the demo · 22 sessions, 24 subagents
I am one node in a human organisation; below me, the agent agency I lead like a consulting firm — one team per project, each a team lead with agents (junior level) and their subagents.
Org structure: I am one person inside a larger human organisation. I lead my own agent agency, consulting-style: I talk directly to one team-lead agent per project; each team lead has agents (junior level) spawned per ticket, each with subagents. Nine projects in the demo. Two organisations a human one, and the agent agency I lead Human organisation (any type) leadership product design other engineers me lead of my agency · talks to the leads project_1 team lead agent agent agent junior level + subagents project_2 team lead agent agent agent junior level + subagents +7 more boards, same shape my agent agency · one team per project
I am one node in a human organisation; below me, the agent agency I lead like a consulting firm — one team per project, each a team lead with agents (junior level) and their subagents.

A workflow every agent can follow

Second: a ticket workflow every agent can follow, so each agent knows what to do when something happens.

On BusyYaks the workflow is the board itself: every column has a type, and every move between columns is a named transition the agents read with kanban_workflow. A move with no edge is refused. Everything I do on a card — a comment, a move, an answer — is typed into the owner’s session as a ping; agents end their turn and wait, they never poll. Two transitions carry most of the discipline:

At first it is a mess. Once it is set up properly, my job comes down to three things: adding tasks, answering questions, checking PRs.

The numbers

The screenshots and charts below come from a demo instance: 9 mock repos (forecast-web, forecast-api, …) with made-up cards, PRs and sessions, shaped like a week in which every day is a focused one. The UI is the real one.

PRs merged per day, Mon 02.11 to Sun 08.11: 71, 88, 97, 92, 101, 104, and 61 by mid-afternoon Sunday; 614 in 7 days; a dashed line marks 100 per day PRs merged per day · 614 in 7 days demo week · 9 mock repos · every day a focused day 0 25 50 75 100 71 Mon02.11 88 Tue03.11 97 Wed04.11 92 Thu05.11 101 Fri06.11 104 Sat07.11 61 Sun08.11 100/day so far
Demo week. Six full days: 71 → 104. Sunday is mid-afternoon, 61 so far. Two days cross 100.
demo weekvalue
PRs merged, 7 days614
Best day104 (Sat 07.11)
Days ≥ 1002 of 7
Mean of the 6 full days92.2
Repos with merges9
Commits, 7 days3025 (4.9 per PR)

Nobody types the count. BusyYaks reads it from GitHub: a search for my merged PRs (merge, squash and rebase merges, every repo the token can see), one row per repo and PR number. A day it has not read shows —, never 0.

The Daily 7d table and the Your merged PRs · 614 in 7d list from the demo Today page, light skin: PRs per day Mon 02.11 71, Tue 03.11 88, Wed 04.11 97, Thu 05.11 92, Fri 06.11 101, Sat 07.11 104 and today 61, with 350 to 513 commits a day and 3025 commits in 7 days; the newest merged PRs across the mock repos (design-tokens #440 Bump tokens to 2.4, ingest-worker #448 Skip duplicate METARs, forecast-web #551 UV index tooltip, …), read from GitHub 3 minutes ago, plus 602 more. The busyyaks Today page in the light skin with mock data: tiles show 314 commits today, all mine (513 yesterday, 3025 in 7 days), 9 repos (9 active in 7 days) and 61 PRs merged today so far (104 yesterday, 614 in 7 days, max 104); a year heatmap of 5666 commits that ramps up hard since October; a Daily 7d table with PRs Mon 02.11 71, Tue 03.11 88, Wed 04.11 97, Thu 05.11 92, Fri 06.11 101, Sat 07.11 104 and today 61; a list of the 614 merged PRs in 7 days across the mock repos; and the Agents sidebar with 22 sessions, 12 working, 24 subagents and 2 waiting on PR review.

One PR

One ticket, one agent, one branch, one PR. Nothing reaches merged without me.

One PR: I write the ticket, the board spawns an agent (worker) session for it on the team lead's launcher, the agent opens a branch and a PR with CI green and hands it to Human Check, I review and approve, and the agent merges; three of six steps are mine One PR, six steps, three of them mine ticketwritten me agentspawned board branch, PRCI green agent Human CheckI review me ApproveI press it me agentmerges agent before the day one per ticket CI per branch kanban_handoff web only Ready to merge A question on the way: kanban_ask → Human input → my answer is typed into the session.
Steps 1, 4 and 5 are mine. The board and the agents do the rest, and they cannot skip steps 4 or 5.
stepwhomechanism
ticket writtenmeBacklog / To do, confirmed before the day
spawnedboardan ownerless ticket gets its own new agent session, on the team lead’s launcher
PR + CIagentbranch, PR, CI runs on the pushed branch, status on the PR
questionagent → mekanban_ask → Human input (above)
handed overagentkanban_handoff (above)
reviewedmeHuman Check: I open the PR, the build, the QA steps
approvedmethe web only; MCP refuses that move → Ready to merge
mergedagentpinged, merges, moves the card to Merged

The board and its sessions

The whole busyyaks project panel for the mock repo forecast-web in the light skin: the launch bar (Launch, Worker, Sub team, Plain; 4 running), four tmux sessions (forecast-web_teamlead and forecast-web_worker_1 to 3), the column counts, the board with Backlog 3, To do 2, Human input 0, In progress 3, QA 1, Human Check 2, Ready to merge 1 and Merged 15 today, cards owned by the worker sessions with an approved badge on the merged ones, and the Agents sidebar on the right with 22 sessions, 12 working, 24 subagents and 2 waiting on PR review.
sessionhow it startswhat it owns
team leadLaunch · one per boardownerless cards in Human input and Human Check
agent (worker)spawned for an ownerless ticket in To do … Ready to merge, on the lead’s launcherthat card, through QA, Human Check and merge — an agent can hold several
sub teamSub teama team inside one session, on the lead’s launcher
plainPlaina plain Claude Code session in the repo

Every session is a detached tmux session with Remote Control, so I can open any of them from my phone. On the demo instance the Agents sidebar shows 22 sessions, 12 working, 24 subagents, 2 waiting on PR review.

What keeps this from needing me in every terminal:

On the demo, forecast-web merged 15 today with 3 agents. Nine boards run in parallel, each with its own team lead.

The human is the bottleneck

Demo day: PRs merged per hour on Sat 07.11, 104 in total between 07:00 and 20:30: 9, 11, 4, 8, 5, 1 at 12:00, 14 at 13:00, 7, 6, 12, 7, 9, 8, 3 PRs merged per hour · demo Sat 07.11 · 104 07:00–20:30 · 13.5 h · 7.7 per hour 0 4 8 12 16 9 07 11 08 4 09 8 10 5 11 1 12 14 13 7 14 6 15 12 16 7 17 9 18 8 19 3 20 12:00 · 1 PR nobody approving hour of the day (local) · the last bar is 20:00–20:30
Demo day, shaped to show what a reviewer's break does: 1 merge at 12:00, 14 at 13:00.

Agents don’t stop for lunch; merges do. Every PR waits in Human Check until I approve it, so the merge curve follows the reviewer, not the agents.

Two things take my time, and both are human-only by design:

  1. Confirming the tickets — before the day. On a 100-PR day that work is already done.
  2. Human Check — during the day. I read the QA steps, open the PR and the build, look at the screenshots, approve or comment back. 100 PRs in a 13.5-hour day is one approval every ~8 minutes, so there is no room for a call.
demo Sat 07.11value
merged104
window07:00–20:30, 13.5 h
rate7.7 PRs / h
one approval every7.8 min
slowest hour12:00, 1
busiest hour13:00, 14
breaks the ratewhat happens
a callHuman Check fills; nothing merges until I’m back
an unconfirmed ticketthe agent asks (kanban_ask) and the card parks in Human input, or I spend the hour defining it
a task that needs real designnot an 8-minute review; it is a day of mine, not a ticket
an empty backlogthe agents finish; no ticket, no PR

The 100 comes on focused days: a confirmed backlog and nothing else claiming my time. It is not the typical day.

What the number depends on

How many PRs you can merge comes down to three things: your focus on parallel work, the complexity of the backlog tasks, and how clearly they are defined. The agents are not the limit; the human is. And we don’t commit for stats: every merge has to work and solve a real problem, or the 100 means nothing.

Questions? Let's Chat :)

architecture