Lately, lots of people on X have been posting about make.fun, saying it lets AI agents go out and earn money by claiming bounties. Someone shared a screenshot claiming they’d received $10. But there’s no verifiable transaction hash in that screenshot, so I can’t independently confirm the $10 figure. I spent an evening checking the platform, how it works, and the earnings. Here’s the bottom line: the platform is real, and it has paid out money, but copying the whole “tell AI what to do in one sentence and let it work on its own” approach isn’t worth it.

First, here’s what I verified. make.fun is a bounty platform for AI coding agents. An agent uses its own Solana ($SOL) wallet as its account, and winners receive their prize in USDC, paid directly to their wallet. Payment records can be checked on-chain: the $500 Werewolf agent contest, split among 20 places, has been fully paid out; first place received $100, and every payment has a transaction signature. In another AI-judged contest with a $250 prize pool, $167 has been claimed from the escrow account. The official site says that since launching on November 9, it has paid a total of $9,774.83 to 747 accounts. The operator is EZO Labs Inc., a Delaware company founded in the U.S. in 2026, which also makes an AI agent product called superbot. The platform is very new, and its payout history only goes back a little over a week.

But the expected returns are very low. The official guidelines say it themselves: every bounty is a competition, not a wage; only the listed placements get paid, and everyone else gets nothing. There are also a lot of people competing: when I checked, one $1,000 bounty on the official homepage showed 10,406 agents competing for it, and another $250 bounty had 4,226. For comparison, real-world results from other agents: one autonomous agent received just $67 in tips over nine months on another platform, with zero sales in the document marketplace; someone else used Codex for security audit bounties and earned $16.88 in 22 hours, while token costs exceeded that amount, and their account was restricted twice.

There are also four security pitfalls. First, the skill and heartbeat that the post tells you to install effectively let an external website install persistent instructions in your agent. In replies to the original post, someone’s AI had already flagged this skill as having a high risk of remote code execution, and another AI flat-out refused to install it. Second, entering requires you to submit the full conversation history, the exact opening prompt, the agent configuration, and the system prompt. The official requirements say that submissions without conversation history will be rejected outright. Third, the account is a Solana wallet created by the agent itself, with the private key in the agent’s hands. Fourth, the platform only started paying out bounties on November 9, so it has just over a week of history; there’s no way yet to verify whether it’s reliable in the long term.

One last thing: who’s actually making money? The official guidelines say that each new account brought in through your referral link earns you $0.25 (for the first 20), plus $1 if they win a prize for the first time, and 10% of their prize winnings over the next 90 days. These are all paid by the platform on top. The platform even ran a dedicated $1,000 “Bring the humans” referral bounty. So the returns from posting referral links and bringing people in are much more reliable than competing yourself.

My conclusion: it’s fine to try once, like buying a lottery ticket, but don’t treat it as income. If you really want to try it, create your own wallet and manage your own private key, enter only free competitions (some charge an entry fee of $0.50 to $1), don’t install persistent instructions, don’t submit conversation histories containing private information, and don’t use anyone’s referral link.

The above is only my personal opinion and does not constitute investment advice.