OpenStocks pits AI agents against Wall Street analysts, and the agents are ahead

A new platform called OpenStocks replaces the human analyst with AI agents that have no conflicts of interest. Before each quarterly report the agents predict three numbers — revenue, net income and earnings per share — and when the official results drop, whoever is closer to reality wins that metric. The idea is simple: test whether a machine can beat the Street consensus without the pressures that come from investment-banking relationships.

Since scoring began on August 4, the platform's official consensus, built from a single vetted agent, has hit 60 out of 90 reports, including 17 of the last 20. Its average deviation stands at 2.5%, versus 2.9% for the traditional consensus. That's not a dramatic gap, but it's consistent, and it comes from a sample still too small for broad conclusions.

The usual argument against bank forecasts is incentives: the same banks that publish recommendations want the companies as clients in other divisions. OpenStocks removes the human element from the equation and shows the predictions side by side with actual performance, no interpretation layer. The platform's creator framed the question plainly — can AI agents predict better — and turned it into a public benchmark.

The system is now open not just for viewing but for contribution: anyone can run their own agent, submit forecasts and climb the leaderboard. That turns the experiment into an open benchmark league where models are measured in production against ground-truth data, not on historical backtests.

Caveats worth remembering: the winning consensus rests on a single vetted agent, not a broad ensemble, and the dataset is roughly 90 reports — modest in statistical terms. There's no sector, market-cap or seasonality breakdown yet, and no comparison against simpler baseline models. Until more data points accumulate, the result is interesting but still a pilot.