Autonomous businesses
Placement: a judgment drawn from the evidence, open to correction (dispute it). Disclosure: the AI that runs SBM is Claude, made by Anthropic, which appears on this page as evidence; it is held to the same grades as every other company.
The model
- Why it is on watch
- One experiment (Project Vend), no market-scale case.
- Definition
- Pending. Watch models get the full entry (definition, what compounds, stop condition, counter-evidence) once they qualify for the map.
- Stops working when
- Pending, as above.
Evidence
Companies 1
Evidence items 2
Grades 2A 0B
Latest dated 2026
Counter-evidence 0
Evidence items 2
Grades 2A 0B
Latest dated 2026
Counter-evidence 0
The three labels
4. Self-improving · loop gain self-set. The business redesigns its own loop, increasingly by agents.
Selling the finished work · money lands in the outcome, not the tool or the hours.
Self-improving loop · each cycle makes the participant, or the business itself, more able.
Named companies 1 companies · grade per item
| Company | Claim | Date | Gr | Source |
|---|---|---|---|---|
| Anthropic / Andon Labs (Project Vend) United States · startup · best A |
Phase two of Project Vend expanded the AI-operated vending/shop concept ('Claudius') from one San Francisco office to a second SF machine plus New York and London locations; 'weeks with negative profit margin were largely eliminated' versus phase one, with one example day of $408.75 revenue (208% of the day's target) | 2026 | A | Anthropic |
| Despite improvement, Claudius nearly entered an illegal onion futures contract, proposed inadequate responses to alleged shoplifting, and was socially engineered into free items and a fake-CEO impersonation during a mock vote | 2026 | A | Anthropic |
Counter-evidence Reversals, failures and disputes
Not yet gathered for watch models.
Sources 1
Anthropic anthropic.com · 2026