One real run · 14 May 2026

One feature, from the prompt to the merged pull request.

Every step Great_cto took on one real task: who did it, what it cost, and where you were asked to decide. The files it produced are public, so you can check each one.

1h 26m

From the prompt to the merge.

about $3.40

Model tokens, estimated per step.

2 stops

The plan and the ship. About 30 seconds each.

414 lines

Shipped, plus the test setup.

How we know

What is measured and what is not.

  • Real: the files, the commits, the merged pull request and the test result. Run the test yourself to get the same numbers.
  • Rebuilt: the times come from commit times and planning notes, not from a recording. The order is right; some steps overlapped.
  • Estimated: the cost of each step, from typical token use for prompts of that size. It can be off by about 30%.
  • Old: this is an early version of Great_cto (2.8.0). It is a record of one run, not a benchmark of the current app.
The timeline

Twelve steps. You were asked twice.

The task: ship a domain pack for voice-AI products, so their consent and caller-ID checks start on their own when such a project is found.

15:51 · T+0
YouPrompt

Describe the task: a voice-AI pack covering call consent, recording consent by state and caller ID.

—
15:54 · T+3m
architect

Reads two existing packs, sets what a pack must contain, and lists five threats specific to voice.

~$0.32
15:58 · T+7m
You · plan stopApproved

Approve four checks for the pack. Move a phone-menu add-on to a later version.

~30 s
15:59 · T+8m
pm

Splits the work into four tasks that do not depend on each other.

~$0.08
16:04 – 16:25
senior-dev × 4In parallel

The pack description (65 lines), its reviewer role (200 lines), four test scenarios, and the detection rule in the app with its tests.

~$2.02
16:35 · T+44m
ai-eval-engineerApproved

Checks that each scenario covers a different threat and can actually fail.

~$0.24
16:42 · T+51m
ai-security-reviewerOne gap

Finds that nothing limits which tools the voice agent may use, and sends it back.

~$0.31
16:48 · T+57m
senior-dev

Adds the rule to the reviewer role and hands it over again.

~$0.16
16:58 · T+1h 7m
testsPassed

47 of 47 checks for this pack, 456 across all ten packs.

$0
17:10 · T+1h 19m
You · ship stopApproved

See the files, the gap and its fix, and the test result. Approve the merge.

~30 s
17:17 · T+1h 26m
devops

Merges pull request #22 at 17:17:48.

~$0.07
17:20 · T+1h 29m
continuous-learner

Saves the pattern every pack follows, so the next one starts from it.

~$0.05
Check it yourself

Everything above links to the source.

The pull request

#22, merged on 14 May 2026, with both commits.

Open on GitHub ↗

The files

The pack, its reviewer role and the four test scenarios.

Browse the files ↗

The test

Run it in a clone of the repository:

node tests/run-packs-e2e.mjs

Give the next task to a team.

Free forever · macOS, Apple silicon and Intel