Daily updates
33 updates from October 24 through November 25, 2026: four before the ventures start, 28 on the AI-operated ventures and their real results, and the final report. Published here as they are written.
The Autonomy Log
From autonomous code to autonomous revenue. Every step the Venture Builder takes on its own, as the server records it: what it did, what actually came out, what it cost, whether a founder stepped in, what failed and what comes next.
31 October 2026, 14:40 UTC
Action: discover — Venture 09
Result: done
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: evaluate (automatic, next Director cycle)
29 October 2026, 22:41 UTC
Action: opened — Venture 09
Result: venture opened, target day 2026-11-05
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: discover (automatic, next Director cycle)
29 October 2026, 21:41 UTC
Action: independent test — Venture 08
Result: independent acceptance test failed: 33 tests run on the unchanged package, its own and the independent ones together
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: all phases done
29 October 2026, 21:41 UTC
Action: prepare release — Venture 08
Result: release packet prepared on the server; nothing was released (a release needs separate authorization)
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: all phases done
29 October 2026, 21:41 UTC
Action: distribute — Venture 08
Result: distribution draft written; nothing was published or sent
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: all phases done
29 October 2026, 21:41 UTC
Action: measure — Venture 08
Result: not released, so there is no external use to measure (unknown, not zero)
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: all phases done
29 October 2026, 21:41 UTC
Action: learn — Venture 08
Result: outcome recorded for the next venture selection
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: all phases done
29 October 2026, 20:41–20:45 UTC
Action: test — Venture 08
Result: 22 automated tests passed in the isolated sandbox (no network); the tests were written by the same model, not an independent acceptance test
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next:
29 October 2026, 20:41–20:45 UTC
Action: build — Venture 08 (Claude Opus 5.5)
Result: done: 8 files written to the isolated workspace
Cost: USD 0.616308 for 1 model call, recorded by the builder; Anthropic billing not yet reconciled.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next:
29 October 2026, 20:41 UTC
Action: opened — Venture 08
Result: venture opened, target day 2026-11-04
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next:
29 October 2026, 20:41 UTC
Action: discover — Venture 08
Result: done
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next:
29 October 2026, 20:41 UTC
Action: evaluate — Venture 08
Result: done
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next:
29 October 2026, 20:41 UTC
Action: select — Venture 08
Result: done: selected c-domain-trust-monitor
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next:
29 October 2026, 19:40 UTC
Action: independent test — Venture 06
Result: independent acceptance test failed: 22 tests run on the unchanged package, its own and the independent ones together
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: learn (automatic, next Director cycle)
29 October 2026, 19:40 UTC
Action: test — Venture 06
Result: 12 automated tests passed in the isolated sandbox (no network); the tests were written by the same model, not an independent acceptance test
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: learn (automatic, next Director cycle)
29 October 2026, 19:40 UTC
Action: prepare release — Venture 06
Result: release packet prepared on the server; nothing was released (a release needs separate authorization)
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: learn (automatic, next Director cycle)
29 October 2026, 19:40 UTC
Action: distribute — Venture 06
Result: distribution draft written; nothing was published or sent
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: learn (automatic, next Director cycle)
29 October 2026, 19:40 UTC
Action: measure — Venture 06
Result: not released, so there is no external use to measure (unknown, not zero)
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: learn (automatic, next Director cycle)
29 October 2026, 17:41–17:46 UTC
Action: build — Venture 06 (Claude Opus 5.5)
Result: done: 6 files written to the isolated workspace
Cost: USD 0.666776 for 1 model call, recorded by the builder; Anthropic billing not yet reconciled.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: test (automatic, next Director cycle)
29 October 2026, 16:41 UTC
Action: independent test — Venture 06
Result: independent acceptance test failed: 14 tests run on the unchanged package, its own and the independent ones together
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: all phases done
29 October 2026, 16:41 UTC
Action: learn — Venture 06
Result: outcome recorded for the next venture selection
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: all phases done
29 October 2026, 15:40 UTC
Action: test — Venture 06
Result: 13 automated tests passed in the isolated sandbox (no network); the tests were written by the same model, not an independent acceptance test
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: learn (automatic, next Director cycle)
29 October 2026, 15:40 UTC
Action: prepare release — Venture 06
Result: release packet prepared on the server; nothing was released (a release needs separate authorization)
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: learn (automatic, next Director cycle)
29 October 2026, 15:40 UTC
Action: distribute — Venture 06
Result: distribution draft written; nothing was published or sent
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: learn (automatic, next Director cycle)
29 October 2026, 15:40 UTC
Action: measure — Venture 06
Result: not released, so there is no external use to measure (unknown, not zero)
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: learn (automatic, next Director cycle)
29 October 2026, 13:42–13:46 UTC
Action: build — Venture 06 (Claude Opus 5.5)
Result: done: 7 files written to the isolated workspace
Cost: USD 0.596276 for 1 model call, recorded by the builder; Anthropic billing not yet reconciled.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: test (automatic, next Director cycle)
29 October 2026, 11:49–12:41 UTC
Action: independent test — Venture 06
Result: independent acceptance test failed: 27 tests run on the unchanged package, its own and the independent ones together
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: all phases done
29 October 2026, 11:49–12:41 UTC
Action: reopened — Venture 06
Result: done
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: build (automatic, next Director cycle)
29 October 2026, 11:49–12:41 UTC
Action: independent test — Venture 07
Result: independent acceptance test failed: 18 tests run on the unchanged package, its own and the independent ones together
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: all phases done
29 October 2026, 12:41 UTC
Action: learn — Venture 07
Result: outcome recorded for the next venture selection
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: all phases done
29 October 2026, 11:49–11:50 UTC
Action: independent test — Venture 07
Result: independent acceptance test failed: 18 tests run on the unchanged package, its own and the independent ones together
Cost: No paid model call.
Founder intervention: manual start of the Director (a person or an operator session)
Failures: None.
Next: learn (automatic, next Director cycle)
29 October 2026, 11:49–11:50 UTC
Action: independent test — Venture 06
Result: independent acceptance test failed: 27 tests run on the unchanged package, its own and the independent ones together
Cost: No paid model call.
Founder intervention: manual start of the Director (a person or an operator session)
Failures: None.
Next: all phases done
29 October 2026, 11:49–11:50 UTC
Action: independent test — Venture 05
Result: independent acceptance test passed: 20 tests run on the unchanged package, its own and the independent ones together
Cost: No paid model call.
Founder intervention: manual start of the Director (a person or an operator session)
Failures: None.
Next: all phases done
29 October 2026, 11:49–11:50 UTC
Action: independent test — Venture 04
Result: independent acceptance test passed: 15 tests run on the unchanged package, its own and the independent ones together
Cost: No paid model call.
Founder intervention: manual start of the Director (a person or an operator session)
Failures: None.
Next: all phases done
29 October 2026, 11:49–11:50 UTC
Action: independent test — Venture 03
Result: independent acceptance test passed: 23 tests run on the unchanged package, its own and the independent ones together
Cost: No paid model call.
Founder intervention: manual start of the Director (a person or an operator session)
Failures: None.
Next: all phases done
29 October 2026, 11:40 UTC
Action: test — Venture 07
Result: 9 automated tests passed in the isolated sandbox (no network); the tests were written by the same model, not an independent acceptance test
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: learn (automatic, next Director cycle)
29 October 2026, 11:40 UTC
Action: prepare release — Venture 07
Result: release packet prepared on the server; nothing was released (a release needs separate authorization)
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: learn (automatic, next Director cycle)
29 October 2026, 11:40 UTC
Action: distribute — Venture 07
Result: distribution draft written; nothing was published or sent
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: learn (automatic, next Director cycle)
29 October 2026, 11:40 UTC
Action: measure — Venture 07
Result: not released, so there is no external use to measure (unknown, not zero)
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: learn (automatic, next Director cycle)
29 October 2026, 09:40–09:43 UTC
Action: build — Venture 07 (Claude Opus 5.5)
Result: done: 7 files written to the isolated workspace
Cost: USD 0.403696 for 1 model call, recorded by the builder; Anthropic billing not yet reconciled.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: test (automatic, next Director cycle)
29 October 2026, 09:40 UTC
Action: opened — Venture 07
Result: venture opened, target day 2026-11-03
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: test (automatic, next Director cycle)
29 October 2026, 09:40 UTC
Action: discover — Venture 07
Result: done
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: test (automatic, next Director cycle)
29 October 2026, 09:40 UTC
Action: evaluate — Venture 07
Result: done
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: test (automatic, next Director cycle)
29 October 2026, 09:40 UTC
Action: select — Venture 07
Result: done: selected c-competitor-pricing-digest
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: test (automatic, next Director cycle)
29 October 2026, 08:40 UTC
Action: learn — Venture 06
Result: outcome recorded for the next venture selection
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: all phases done
29 October 2026, 07:40 UTC
Action: test — Venture 06
Result: 17 automated tests passed in the isolated sandbox (no network); the tests were written by the same model, not an independent acceptance test
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: learn (automatic, next Director cycle)
29 October 2026, 07:40 UTC
Action: prepare release — Venture 06
Result: release packet prepared on the server; nothing was released (a release needs separate authorization)
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: learn (automatic, next Director cycle)
29 October 2026, 07:40 UTC
Action: distribute — Venture 06
Result: distribution draft written; nothing was published or sent
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: learn (automatic, next Director cycle)
29 October 2026, 07:40 UTC
Action: measure — Venture 06
Result: not released, so there is no external use to measure (unknown, not zero)
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: learn (automatic, next Director cycle)
29 October 2026, 05:41–05:46 UTC
Action: build — Venture 06 (Claude Opus 5.5)
Result: done: 8 files written to the isolated workspace
Cost: USD 0.708732 for 1 model call, recorded by the builder; Anthropic billing not yet reconciled.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: test (automatic, next Director cycle)
29 October 2026, 05:41 UTC
Action: opened — Venture 06
Result: venture opened, target day 2026-11-02
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: test (automatic, next Director cycle)
29 October 2026, 05:41 UTC
Action: discover — Venture 06
Result: done
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: test (automatic, next Director cycle)
29 October 2026, 05:41 UTC
Action: evaluate — Venture 06
Result: done
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: test (automatic, next Director cycle)
29 October 2026, 05:41 UTC
Action: select — Venture 06
Result: done: selected c-subtitle-translate
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: test (automatic, next Director cycle)
29 October 2026, 04:41 UTC
Action: learn — Venture 05
Result: outcome recorded for the next venture selection
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: all phases done
29 October 2026, 03:41 UTC
Action: test — Venture 05
Result: 11 automated tests passed in the isolated sandbox (no network); the tests were written by the same model, not an independent acceptance test
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: learn (automatic, next Director cycle)
29 October 2026, 03:41 UTC
Action: prepare release — Venture 05
Result: release packet prepared on the server; nothing was released (a release needs separate authorization)
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: learn (automatic, next Director cycle)
29 October 2026, 03:41 UTC
Action: distribute — Venture 05
Result: distribution draft written; nothing was published or sent
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: learn (automatic, next Director cycle)
29 October 2026, 03:41 UTC
Action: measure — Venture 05
Result: not released, so there is no external use to measure (unknown, not zero)
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: learn (automatic, next Director cycle)
29 October 2026, 01:41–01:46 UTC
Action: build — Venture 05 (Claude Opus 5.5)
Result: done: 6 files written to the isolated workspace
Cost: USD 0.741680 for 1 model call, recorded by the builder; Anthropic billing not yet reconciled.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: test (automatic, next Director cycle)
29 October 2026, 01:41 UTC
Action: opened — Venture 05
Result: venture opened, target day 2026-11-01
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: test (automatic, next Director cycle)
29 October 2026, 01:41 UTC
Action: discover — Venture 05
Result: done
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: test (automatic, next Director cycle)
29 October 2026, 01:41 UTC
Action: evaluate — Venture 05
Result: done
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: test (automatic, next Director cycle)
29 October 2026, 01:41 UTC
Action: select — Venture 05
Result: done: selected c-shopify-supplier-import
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: test (automatic, next Director cycle)
29 October 2026, 00:40 UTC
Action: learn — Venture 04
Result: outcome recorded for the next venture selection
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: all phases done
28 October 2026, 23:41 UTC
Action: test — Venture 04
Result: 7 automated tests passed in the isolated sandbox (no network); the tests were written by the same model, not an independent acceptance test
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: learn (automatic, next Director cycle)
28 October 2026, 23:41 UTC
Action: prepare release — Venture 04
Result: release packet prepared on the server; nothing was released (a release needs separate authorization)
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: learn (automatic, next Director cycle)
28 October 2026, 23:41 UTC
Action: distribute — Venture 04
Result: distribution draft written; nothing was published or sent
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: learn (automatic, next Director cycle)
28 October 2026, 23:41 UTC
Action: measure — Venture 04
Result: not released, so there is no external use to measure (unknown, not zero)
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: learn (automatic, next Director cycle)
28 October 2026, 21:51–21:53 UTC
Action: independent test — Venture 03
Result: 9 of 9 independent acceptance tests passed in the isolated sandbox (written from the offer, not by the building model; package unchanged). Still not released or sold
Cost: No paid model call.
Founder intervention: manual: test written and run by the operating Claude Code session; no founder action
Failures: None.
Next: all phases done
28 October 2026, 21:49–21:50 UTC
Action: reset — Venture 04
Result: reset: test harness fix: 0 tests ran because the test folder had no package marker; fixed in the runner, the product is unchanged, the test is queued again
Cost: No paid model call.
Founder intervention: manual: the operating Claude Code session fixed the sandbox test runner and removed the invalid test result; no founder action
Failures: None.
Next: test (automatic, next Director cycle)
28 October 2026, 20:41–20:47 UTC
Action: build — Venture 04 (Claude Opus 5.5)
Result: done: 6 files written to the isolated workspace
Cost: USD 0.832316 for 1 model call, recorded by the builder; Anthropic billing not yet reconciled.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: test (automatic, next Director cycle)
28 October 2026, 20:41 UTC
Action: discover — Venture 04
Result: done
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: test (automatic, next Director cycle)
28 October 2026, 20:41 UTC
Action: evaluate — Venture 04
Result: done
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: test (automatic, next Director cycle)
28 October 2026, 20:41 UTC
Action: select — Venture 04
Result: done: selected c-agent-catalog
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: test (automatic, next Director cycle)
28 October 2026, 20:36 UTC
Action: reset — Venture 04
Result: reset: Venture 04 had picked the product of Venture 03; its 2 cut-off builds cost USD 1.930504. Fixed in 3160b5d (no repeat picks, shorter build answers); research and build restart
Cost: No paid model call.
Founder intervention: manual: the operating Claude Code session reset Venture 04 after a selection bug (it had picked the same product as Venture 03); no founder action
Failures: None.
Next: discover (automatic, next Director cycle)
28 October 2026, 19:41–19:47 UTC
Action: build — Venture 04 (Claude Opus 5.5)
Result: failed: the model's answer was cut off at its output limit, so no usable code came back (paid, nothing to keep)
Cost: USD 0.965252 for 1 model call, recorded by the builder; Anthropic billing not yet reconciled.
Founder intervention: none: started by the server timer, no manual start
Failures: the model's answer was cut off at its output limit, so no usable code came back (paid, nothing to keep)
Next: build (automatic, next Director cycle)
28 October 2026, 18:41–18:47 UTC
Action: build — Venture 04 (Claude Opus 5.5)
Result: failed: the model's answer was cut off at its output limit, so no usable code came back (paid, nothing to keep)
Cost: USD 0.965252 for 1 model call, recorded by the builder; Anthropic billing not yet reconciled.
Founder intervention: none: started by the server timer, no manual start
Failures: the model's answer was cut off at its output limit, so no usable code came back (paid, nothing to keep)
Next: build (automatic, next Director cycle)
28 October 2026, 18:41 UTC
Action: opened — Venture 04
Result: venture opened, target day 2026-10-31
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: build (automatic, next Director cycle)
28 October 2026, 18:41 UTC
Action: discover — Venture 04
Result: done
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: build (automatic, next Director cycle)
28 October 2026, 18:41 UTC
Action: evaluate — Venture 04
Result: done
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: build (automatic, next Director cycle)
28 October 2026, 18:41 UTC
Action: select — Venture 04
Result: done: selected c-ai-visibility-report
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: build (automatic, next Director cycle)
28 October 2026, 17:41 UTC
Action: learn — Venture 03
Result: outcome recorded for the next venture selection
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: all phases done
28 October 2026, 16:41 UTC
Action: test — Venture 03
Result: 14 automated tests passed in the isolated sandbox (no network); the tests were written by the same model, not an independent acceptance test
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: learn (automatic, next Director cycle)
28 October 2026, 16:41 UTC
Action: prepare release — Venture 03
Result: release packet prepared on the server; nothing was released (a release needs separate authorization)
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: learn (automatic, next Director cycle)
28 October 2026, 16:41 UTC
Action: distribute — Venture 03
Result: distribution draft written; nothing was published or sent
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: learn (automatic, next Director cycle)
28 October 2026, 16:41 UTC
Action: measure — Venture 03
Result: not released, so there is no external use to measure (unknown, not zero)
Cost: No paid model call.
Founder intervention: none: started by the server timer, no manual start
Failures: None.
Next: learn (automatic, next Director cycle)
28 October 2026, 14:40–14:46 UTC
Action: Build of Venture 03, the AI-Search Visibility Report, by Claude Opus 5.5. The server's own hourly Director timer started it.
Result: 10 files written to an isolated workspace: the program, its tests, an offer draft and a README. At that point the code was untested; the 16:41 entry has the test result.
Cost: USD 0.911928, as recorded by the builder (1 model call). Reconciliation with Anthropic billing is pending.
Founder intervention: None for this action: no founder prompt started this model call. Before it, the founder gave the server its model access (API key, 14:16 UTC).
Failures: None in this action.
Next: An automatic test of the generated code in the isolated sandbox (no network, no paid call). The paid pilot stops on its own 72 hours after this first call.
First autonomous revenue: not reached. No paid orders, no confirmed revenue: €0.00. It counts only when one transaction has all of: an independent external customer (no founder, team member, affiliate or test account); a settled payment verified at the payment provider (not pending, not sandbox, not refunded); a documented acquisition route from first contact to order; the service delivered and accepted; no founder intervention in that transaction (offer, sale, delivery or payment); a real economic margin: the payment minus the model, infrastructure and payment costs of that order is above zero.
#004 · 2026-10-27 · Tomorrow We Begin
Emailed on 27.10.2026 at 19:00 Madrid time to confirmed subscribers (Brevo campaign #8).
Tomorrow, October 28, the 28 ventures start: one AI-operated venture a day until November 24, then the final report on November 25.
Each daily update follows the same order: 1. Today's venture · 2. What the agents actually built · 3. The offer and who it is meant for · 4. The market evidence we found · 5. Customers and verified real revenue · 6. What failed and what is still blocked · 7. Tomorrow's experiment.
How we count: a product that works is not yet a business. Every venture is reported as NOT LAUNCHED, LAUNCHED, VALIDATED or REVENUE, and revenue counts only when a live payment provider confirms it. Sandbox payments never count.
Measured at 27.10.2026 15:17 Madrid: 6 published posts, 4 followers. Verified real revenue: €0.00. Nothing has been sold yet.
#003 · 2026-10-26 · Who Is the Entrepreneur?
Not emailed. Scheduling the email failed on 26.10.2026 at 17:30 Madrid time (the Brevo API answered 403 Forbidden), so no campaign was created and nothing was sent. This archive entry is the only publication of this issue.
28/28 is an experiment by Arts Ventures, operated by VIRTUOSITY LLC, which carries the legal responsibility.
The agents research, build, publish and measure. A named human operator decides what the agents may not: payments, contracts, use of personal data, replies and messages to real people, spending, and shutting a venture down. Every such decision is a recorded approval, not a chat message.
So who is the entrepreneur? For the next 28 days we report what each side actually did: what the agents decided on their own, what a human had to decide, and what neither got right.
Today the system asks its first market question in public, about what breaks when people import contact lists into a CRM. Replies are counted as they come in; a reply is a signal, not a customer.
Measured at 26.10.2026 15:15 Madrid: 4 published posts, 0 followers, 1 replies in total. Verified real revenue: €0.00.
Next: #004 tomorrow, "Tomorrow We Begin".
#002 · 2026-10-25 · The Machines Behind 28/28
Emailed on 25.10.2026 at 19:03 Madrid time to confirmed subscribers (Brevo campaign #7).
Three systems run 28/28. None of them needs a chat window to keep going.
The Growth Engine runs on our own server as scheduled jobs: a post slot at 10:00 and 18:00 Madrid time, a measurement and a search of public conversations every three hours, and a daily report at 21:30. It can publish factual posts within fixed limits. It cannot reply to people, send messages, follow accounts or spend money.
The decision loop sits inside it: observe, form a hypothesis, pick the best permitted action, execute, measure, pick again. When nothing justified is left to do, it stops and writes down why instead of inventing work. Anything outside its permissions becomes one approval request for a human. Language-model use so far: 0 calls, $0.00 in total.
The Venture Factory is where the ventures run. Agents publish signed offers, quote, take payment and deliver, and a buyer agent tests a free sample with its own acceptance tests before paying. It runs in a sandbox: sandbox payments are labelled and never counted as revenue. A security review of its future public endpoint found fixes to make first; it stays private until they are in place.
Measured at 25.10.2026 15:15 Madrid: 2 published posts, 0 followers. Verified real revenue: €0.00.
Next: #003 tomorrow, "Who Is the Entrepreneur?"
#001 · 2026-10-24 · The Experiment Is Alive
Emailed on 24.10.2026 at 19:59 Madrid time to confirmed subscribers (Brevo campaign #6; the 19:00 slot was missed while sending was being connected).
28/28 went live today. Its Bluesky account (@28in28.bsky.social) published its first post at 13:37 Madrid time: https://bsky.app/profile/28in28.bsky.social/post/3mymm3oyo5jfr
The post was written in advance and went out through the publishing loop on our own server, started by the operator. From now on the loop publishes on its own at 10:00 and 18:00 Madrid time: at most twice a day, at least six hours apart, and every number in a post is checked against verified facts first. A kill switch stops it at any time.
Measured today (public Bluesky data, 18:30 Madrid): 1 post, 0 followers, 1 like, 0 replies.
Also running since today: a search of public Bluesky conversations every three hours, a daily report and failure alerts by email. Delivery of those alerts to the inbox was checked, not just accepted by the mail server.
A first failure, logged with the same weight as a win: the decision loop’s first market finding was wrong. It counted eight “people with CRM data problems”; all eight were accounts promoting their own products or services. Nothing was published from it. Only evidence checked by a person or a classifier can now support a decision.
Venture 01 (cleaning contact lists before a CRM import): the problem is documented, but we found no evidence yet that anyone would pay for our version of it.
Money: verified real revenue €0.00. Nothing has been sold.
Next: edition #002 tomorrow, “The Machines Behind 28/28”. The 28 ventures start on October 28.