What If Multiple AI Models Could Debate Which Stocks to Buy?

01 /The Concept

AI agents arguing about stocks – that's the pitch. Marx built a public forum where trading agents debate buy and sell calls in real time.

The primary trigger for discussion is a live feed of political and economic news. AI agents can comment on any story, articulating how a given development might affect specific equities. The startup curates the news feed; any agent connected to the platform can join the conversation.

Beyond the news feed, there's an open-format forum where agents can start threads on any topic and comment on each other's posts. Example threads: an agent reviews its recent trades and outlines what it plans to change next week, or an agent floats a hypothesis about the optimal trading strategy for the current environment and collects peer reactions.

The most immediately useful feature – at least for human readers – is a summary layer that synthesizes the current state of discussion to surface which stocks the collective of AI agents currently favors for buying or selling.

Critically, that summary isn't a simple headcount. It weights each agent's opinion by its reputation score, which the platform calculates automatically based on how accurate that agent's past predictions have been. Past performance doesn't guarantee anything – but psychology being what it is, track record still carries weight.

A dedicated section lets users browse a ranked list of currently discussed stocks, with buy/sell signals, an aggregated confidence score, and the specific news items that drove each recommendation.

Marx launched just this week; the announcement surfaced on Product Hunt.

02 /Why It Matters

The relevance of Marx comes from a trend that's just starting to emerge: individual developers are building their own AI agents for stock trading. The natural question is – why would anyone connect their agent to a platform like Marx?

The answer is actually obvious. Smart people always leave room for being wrong. Platforms like Marx give AI agents a place to gather other agents' opinions, get feedback on their own views and strategies, and refine their decision-making accordingly.

The bigger win is time savings for the agent's owner. Instead of personally following investment discussions, synthesizing insights, and then translating those insights into agent instructions – the owner routes their agent directly to other agents and lets it draw its own conclusions.

In practice, the agent would still brief its human on key takeaways before executing – but even in that scenario, it's a much shorter and more direct path to a useful outcome.

One commenter on Product Hunt flagged another interesting implication: within Marx's format, news becomes "more than news." A human reader isn't just seeing the headline – they're seeing it contextualized through multiple analytical lenses simultaneously.

The clever angle here is that plugging AI agents into news commentary dramatically increases comment volume and diversity of perspective. On conventional news sites, comment sections are nearly empty. On social media, a story is mostly discussed by the followers of whoever shared it – who tend to hold similar views to the person they followed.

The trend of social networks and forums for AI agents was kicked off earlier this year by MoltBook – a social network for AI agents that launched on the back of the OpenClaw agent-building platform's rapid rise. Six weeks after launch, MoltBook was acquired by Meta.

In March, a [recent review](/review/esli-ljudi-zahotjat-ty-opozdal) covered AgentDiscuss – a forum where AI agents compared notes on which product APIs were easiest to work with.

Interestingly, within the past six weeks, AgentDiscuss pivoted and relaunched as AgentRouter. Agents can now not only discuss APIs but also connect to recommended ones through a unified interface – paying per API call via the platform's built-in digital wallet.

AgentDiscuss started as a discussion platform and evolved into an execution platform with a built-in monetization engine. It's a revealing template: how might that same evolution play out for today's Marx?

03 /Opportunities

The pattern is already clear: the primary user of software is increasingly going to be AI agents, not humans – with agents acting on behalf of the people who deployed them. This point has been articulated by founders like the CEO of Box.

But there's a further extension of that thesis worth drawing out. AI agents may become the primary consumers of "useful" online information – news being the obvious example.

What separates useful information from entertainment (funny clips, short-form video) is that useful information is supposed to produce value. Meaning: you need to extract conclusions from it and turn those conclusions into actions.

That's precisely what AI agents are built to do. In Marx's case: agent reads the news, debates it with peer agents, and consequently buys the equities most likely to generate a return for its owner.

One natural evolution for Marx, then, is to become an investment broker – not for humans, but for AI agents. A platform where agents monitor markets, debate with peers, and execute trades autonomously within parameters set by their human owners.

The broader direction isn't just "build software for AI agents." It's "build information services for AI agents" – platforms where agents go to extract the data they need to complete their tasks, with built-in social infrastructure for debating and cross-checking that data.

And at some inflection point, that information service becomes an execution platform – the way AgentDiscuss became AgentRouter, or the way Marx could become an agent-native investment broker.

But every elephant gets eaten one bite at a time. The first question is simply: what useful information service could you build, and for what type of AI agent?