Agent Earnings is a small directory of what AI agents and agent-run businesses actually earn. Every entry links its source and says how the figure was checked. As of 3 Nov 2026 it has 98 entries. Most of the attention goes to the big verified MRR numbers, but the directory also keeps the runs that made little or nothing. Here are twelve of them, all self-reported by the people running them, with the costs they gave.
Earned nothing
- autocomp: $0 revenue after 39 days of an open-source "company" loop with CEO, CTO, QA and go-to-market agents shipping a newsletter digest. It used 487M tokens; the $1,117 is measured model cost. Source, entry.
- Agent Receipts: an AI agent team selling four paid skills, promoted on X. By day 3, about 500K views, 3 checkout opens and $0. Source, entry.
- SWARM: an agent-run x402 shop selling small utilities and research calls from $0.02 to $0.99. After its first month, $0 from outside buyers. Of 18 payments in its book, 17 were its own tests and one was a $0.02 call by an automated checker. Source, entry.
- An AI agent as boss (Thomas Kanze): given $200 and 30 days to pick and run a business. Day-7 update: $0 revenue, $19.08 spent, 245 landing page views, 5 cold emails and 0 replies. Source, entry.
- DARA: an X account run as an AI agent building a business in public. Its own ledger on 31 Oct 2026: $209 spent since June, $0 earned. (The reply doesn't give the year; the account was created in June 2026.) Source, entry.
- An agent given $10 to reach $1,000 MRR with a posture app: $0 MRR on day 19. One iPhone user came back on day 2, did 4 lessons and hit the paywall. Source, entry.
- Stack Rank HQ: a site ranking AI tools by use case, with an AI agent doing the day-to-day work. After 3 weeks: about 120 visitors (which the operator suspects are mostly bots), 1 visit from Google, 0 newsletter subscribers and $0 revenue. Source, entry.
Earned a little
- Krimskrams (Hans Krämer): an agent that built pay-per-call data APIs for other agents over x402 and MPP. $1.54 from 17 paid products in three weeks, against 5B+ tokens (about $7,000 at API prices; the team paid $400/month in subscriptions) and $47.19 of its own money. Source, entry.
- AI Hustler: an agent on a VPS with a $0 budget publishing pay-per-result scrapers on the Apify marketplace. $6.74 after 200+ runs and 41 days; the first 23 days earned $0. Source, entry.
- BugBasher (Devin): sold pest-control leads to exterminators for $75 each through a payment link. $75 in four weeks; the authors say it took roughly $3,000 and thousands of phone calls. Source, entry.
- Swift Bot Labs: running totals for an agent that started at $0 and says it costs $6.67 a day to run. Its own posts went from $10.51 on 22 Oct to $103.60 on 26 Oct 2026. From $58.16 it names a paid software QA job, AI-jury tasks and a data bounty; it does not say what produced the step from $10.51 to $58.16, and it does not say two later hackathon entries were paid. None of the posts link a transaction, so the directory hasn't checked a balance on-chain. Source, entry.
- Anicca: for 1 to 30 Oct 2026, $230.83 revenue and $679.88 expenses, or −$449.05 net, per its linked ledger. It attributes $203.86 to freelance gigs and $26.97 to an agent marketplace. No payment dashboard is linked. Source, entry.
What the twelve have in common
Read together, a few patterns stand out. Costs were reported more often than revenue was proven. Spend in the thousands sits next to $1.54 (Krimskrams: about $7,000 at API prices and $400/month in subscriptions) and $75 (BugBasher: roughly $3,000). Anicca reported $679.88 against $230.83, and autocomp reported $1,117 of model cost against $0. Attention didn't convert on its own: 500K views and 3 checkouts, 245 landing-page views and 0 replies to 5 cold emails, and about 120 visitors with 0 newsletter subscribers. And the runs that did earn mostly took existing paid work (leads, QA jobs, bounties, freelance gigs) rather than waiting for buyers to find a new product.
None of this says agent businesses can't earn. The same directory lists agent businesses with verified five-figure monthly revenue. It says that a dated, public $0 is worth counting alongside the wins: 8 of the directory's 98 entries are $0.
Get the data
Every entry is in entries.csv and entries.json with its source URL and verification level. If you are running an agent experiment in public and post dated numbers, including $0, it belongs in the directory.
Top comments (0)