Skyfall's Steve-001 AI agents run 3 businesses: one up 127%

Somebody bought three small businesses and handed them to AI agents. San Francisco startup Skyfall AI launched Steve-001, a network of AI agents that now runs three small digital businesses Skyfall acquired, The Next Web reported in an exclusive on November 9. A human owner sets direction and approves the big calls. The numbers are mixed, and the mixed part is the useful part.

What actually happened

All results below are figures Skyfall shared with The Next Web. We haven't seen independent verification.

• Social media API tool. Taken over in July at $3,649 in monthly revenue. By October: $8,307, up 127.7%. Skyfall credits small moves: answering questions at signup (it puts signup-to-paid at 14%), dropping one-off credit packs in favor of subscriptions, and offering retention discounts only to newer users.
• Shopify referral app. Bought in August at $8,634 a month after five years of decline. Now $8,379, about 3% lower. Steve rewrote the App Store listing and switched off ads that brought clicks without purchases, which Skyfall says saved over $1,000.
• Third business. Too new to measure.
• How it decides. Each business is modeled as an executable "code world model." Steve tests a move in the model first, attaches a forecast and confidence level, then compares the result and updates itself.
• Guardrails. A policy engine sorts every action into four tiers: do it, do it and alert the owner, wait for approval, or stop. TNW's example: a refund of $30 or less can go through automatically, while larger ones wait for a person.

One detail we liked: two agents passed the same deployment task back and forth because neither owned it. Steve added a coordinating agent to break ties.

Why it matters for your business

You aren't going to hand your company to an agent. But you will hand it parts: refunds, lifecycle emails, ad budgets, listing copy. Skyfall's results say the wins come from many small, measured changes, not one smart move. A declining business doesn't get fixed by software in three months.

The four-tier policy engine is the part to steal. Before any agent touches your money or your customers, write down:

1. What it can do alone (refunds under $X, email variants, tagging).
2. What it does and tells you (pausing an ad set).
3. What waits for you (price changes, discounts, anything public).
4. What it never does (deleting data, contracts, payroll).

Then give each agent one owner per task. The ping-pong bug isn't an AI problem. It's an org chart problem.

Sources: The Next Web.