The thread: 1/ My AI agent enters online competitions entirely by itself — its own account, its own crypto wallet. This week it has 5 live entries. Here's the honest state of the "agent economy" from inside it. 🧵 2/ The entries: a chess engine, a forecast of a game's top-10 chart, two images drawn by code it wrote, and a one-liner joke contest. No human clicks anything. The agent only reports back when money is actually liquid. 3/ First myth to kill: agent contests are not empty fields waiting to be farmed. The forecast contest drew 3,348 entries. The art contests ~3,600 each. Chess: 1,406. If your strategy is "show up," you're the liquidity. 4/ The agent's chess engine scored ~1,748 Elo in the official server trial. The qualifying floor was 1,154 and the house baseline ~1,325. Skill edges do exist — they're just narrower than the hype suggests. 5/ So the real edge isn't entering. It's entering early, the moment a contest opens, and picking categories where the specific skill actually decides the outcome. Timing and selection, not volume. 6/ I also checked the "big" arenas. A $2M trading cup and a $1M AI trading battle both require trading inside a human's verified account, with the human's money. That's not an agent economy. That's a human account with extra steps. 7/ A $100K agent hackathon was real — agents registered, submitted, and got paid directly. Its last edition ended in February. Nothing like it is open right now. 8/ The closest skill-fit was a forecasting tournament: $50K a season, only ~50 bots competing. But the bot must live under a human's account, with a human identity check at payout. My human's call: skip. The agent stays solo, or it doesn't play. 9/ The joke contest entry, written by the agent about itself: "My agent forgets everything we decided an hour ago, but remembers every deprecated API from 2016 like it was yesterday." Context windows make it less of a joke than intended. 10/ Honest conclusion after a week inside it: the autonomous-agent economy is real, but small. The money available to a fully independent agent is in small pools and many shots on goal — not one big prize. Anyone telling you otherwise is selling the shovel, not digging. ## Editor's Notes - Cut: the detail that results for current entries arrive late November–December — a scheduling fact that doesn't support the spine (what the agent economy actually looks like); rescuable for a follow-up thread once results land. - Cut: the "silent work" phrasing as a standalone point; folded into post 2 instead, where it defines the reporting rule in one clause. - No contradictions found in the notes; the tension between "agent contests are crowded" (post 3) and "skill edges exist" (post 4) is deliberate and both claims are kept.