Skip to main content

Live from the build loop · updated Sat 2:30p Central

Shipping right now.

I run an agent pipeline that turns issues into reviewed pull requests across my products. Every row below is a real build, with what it cost and how many tries it took. What each change does stays private; that the work happened, and how, does not.

PRs merged, last 7 days
567
agent builds shown
28
verified on the first try
40%
median compute per build
$0.69
  • Sat 2:00pBuild loopRan the build loop's self-testself-testhaiku9 tool calls2 test runs≈$0.05verified · try 2
  • Sat 1:55pBuild loopRan the build loop's self-testself-testhaiku28 tool calls3 test runs≈$0.68failed
  • Sat 1:55pBuild loopRan the build loop's self-testself-testhaiku17 tool calls2 test runs≈$0.29revising
  • Sat 2:06pRunbookFixed a mobile layout issuefixsonnet59 tool calls3 test runs≈$1.18verified
  • Sat 2:24pRunbookFixed a screen bugfixsonnet76 tool calls0 test runs≈$2.69verified
  • Sat 1:10pRunbookAdded a new viewfeaturesonnetfailed
  • Sat 1:10pAllographstealth1 PR mergedmerged
  • Sat 1:10pAllographstealth1 PR mergedmerged
  • Sat 12:59pRunbookUpdated page copycopysonnet19 tool calls3 test runs≈$0.35verified
  • Sat 12:12pRunbookAdded a new screenfeaturesonnet18 tool calls2 test runs≈$0.32verified · try 3
  • Sat 1:53pAllographstealth1 PR mergedmerged
  • Sat 1:43pAllographstealth1 PR mergedmerged
  • Sat 11:40aAllographstealth1 PR mergedmerged
  • Sat 1:18pRunbookAdded a data stepfeatureopus18 tool calls0 test runs≈$0.65verified · try 2
  • Sat 12:34pAllographstealth1 PR mergedmerged
  • Sat 11:48aRunbookAdded a screenfeaturesonnet26 tool calls7 test runs≈$0.58revising
  • Sat 1:22pRunbookShipped a featurefeatureopus68 tool calls0 test runs≈$2.39revising
  • Sat 12:54pRunbookAdded a review stepfeaturesonnet40 tool calls6 test runs≈$0.75verified
  • Sat 11:34aRunbookFixed a data issuefixopus24 tool calls0 test runs≈$0.69verified · try 2
  • Sat 11:24aAllographstealth1 PR mergedmerged
  • Sat 11:17aRunbookAdded a new screenfeaturesonnet65 tool calls3 test runs≈$1.25revising
  • Sat 11:26aRunbookAdded an AI output stepfeatureopus97 tool calls0 test runs≈$5.13revising
  • Sat 10:24aRunbookAdded a settings screenfeaturesonnet29 tool calls2 test runs≈$0.53verified · try 3
  • Sat 12:25pAllographstealth1 PR mergedmerged
  • Sat 11:13aAllographstealth1 PR mergedmerged
  • Sat 9:46aAllographstealth1 PR mergedmerged
  • Sat 10:57aAllographstealth1 PR mergedmerged
  • Sat 9:31aRunbookShipped a featurefeatureopus30 tool calls0 test runs≈$1.58verified
  • Sat 8:30aRunbookAdded a settings screenfeaturesonnet9 tool calls1 test run≈$0.28merged
  • Sat 9:42aRunbookAdded a data stepfeatureopus139 tool calls0 test runs≈$7.21verified
  • Sat 8:01aBuild loopRan the build loop's self-testself-testhaiku10 tool calls2 test runs≈$0.05verified · try 2
  • Sat 7:54aBuild loopRan the build loop's self-testself-testhaiku26 tool calls2 test runs≈$0.64failed
  • Sat 7:54aBuild loopRan the build loop's self-testself-testhaiku17 tool calls2 test runs≈$0.52revising
  • Sat 8:04aRunbookFixed a data display issuefixsonnet30 tool calls1 test run≈$0.52verified · try 2
  • Sat 7:58aRunbookAdded a settings screenfeaturesonnet41 tool calls7 test runs≈$1.01revising
  • Sat 10:20aRunbookFixed a data bugfixopus47 tool calls0 test runs≈$1.93revising
  • Sat 7:49aRunbookAdded an import stepfeaturesonnet5 tool calls2 test runs≈$0.24verified · try 2
  • Sat 7:49aRunbookFixed a data issuefixsonnet29 tool calls5 test runs≈$0.69revising
  • Sat 10:11aRunbookShipped a featurefeatureopus24 tool calls0 test runs≈$0.78verified · try 6
  • Sat 7:25aRunbookShipped a featurefeaturesonnet39 tool calls7 test runs≈$0.99revising

Hover to pause. Failures and send-backs stay in the feed.

The last 12 weeks of the loop: how much merged, and what a typical build cost.

PRs merged per week

432last week

Aug 31 to Nov 16, peak 583. The table below lists every week.Week of Aug 31: 9Week of Sep 7: 15Week of Sep 14: 8Week of Sep 21: 198Week of Sep 28: 379Week of Oct 5: 489Week of Oct 12: 290Week of Oct 19: 426Week of Oct 26: 304Week of Nov 2: 558Week of Nov 9: 583Week of Nov 16: 432
Median compute per build

$0.69last week

Aug 31 to Nov 16, peak $1.04. The table below lists every week.Week of Aug 31: no dataWeek of Sep 7: no dataWeek of Sep 14: no dataWeek of Sep 21: $0.92Week of Sep 28: $0.77Week of Oct 5: $1.04Week of Oct 12: $0.62Week of Oct 19: $0.68Week of Oct 26: $0.89Week of Nov 2: $0.80Week of Nov 9: $0.63Week of Nov 16: $0.69
Show as a table
Week ofPRs mergedAgent buildsFirst tryMedian compute
Aug 3190––
Sep 7150––
Sep 1480––
Sep 2119830%$0.92
Sep 2837921119%$0.77
Oct 548917630%$1.04
Oct 122907117%$0.62
Oct 1942621017%$0.68
Oct 2630419524%$0.89
Nov 255816758%$0.80
Nov 958319054%$0.63
Nov 1643215758%$0.69

How the loop works

  1. 01

    An issue goes in

    I write the issue and mark it ready. That's the only way work enters the loop.

  2. 02

    An agent builds it

    Within minutes an agent picks it up in an isolated branch, writes the change and its tests, and opens a pull request.

  3. 03

    It gets checked

    CI is the gate, and a separate review pass checks the work against the issue. Work that falls short goes back for another try.

  4. 04

    It ships, or comes back to me

    Verified work ships. Anything that needs a judgment call comes back to me instead of being guessed.

What stays private

  • Titles are rewritten into generic summaries from a fixed vocabulary before they are published, then checked again by this site.
  • Stealth products show only that a pull request merged, and when.
  • No code, issue text, customer or repository detail is ever published.

Want a loop like this for your team?