Cheaper agents change the equilibrium price of research

Cheaper research agents should change how much markets pay for ordinary information processing. Our 2026–2031 scenario is that wider adoption reduces the persistence of some public-information opportunities while moving research effort toward harder questions. More participants can investigate the same event, and the remaining reward must cover a different cost structure. This is an equilibrium projection: adoption changes the opportunity being evaluated.

The relevant horizon is hours to days, which we call midhorizon here. A public announcement may contain information whose implications take time to interpret. Agents that lower interpretation costs could bring those implications into prices sooner. Faster adjustment can benefit price discovery while reducing the returns available to participants who previously earned money performing that interpretation.

Grossman and Stiglitz's 1980 paper supplies the theoretical starting point. In their model, costly information acquisition and imperfectly revealing prices interact: informed participants require compensation for collecting information. This is a model of incentives, rather than a finding about present-day agents. Lower research costs change those incentives; they leave room for further costs and uncertainty.

A cheaper question changes participation

Imagine an illustrative signal available in a public release. A small group previously spent substantial time turning it into an instrument-level forecast. If agents perform that routine synthesis cheaply, more participants can inspect the signal. Orders responding to it can move the price before the previous holding window ends. The old backtest then describes a market with less competition than the new one.

The mechanism requires useful interpretation and feasible trading. Agents that repeat headlines without improving forecasts may increase message volume while leaving prices little changed. Agents that agree on a forecast but face prohibitive execution costs may also have limited impact. We expect the strongest adjustment where public evidence, reliable interpretation and economical market access meet.

Research labor is only part of information cost. Fresh data can be expensive to obtain or validate. An unusual instrument may require specialized knowledge. Testing a hypothesis consumes compute and researcher attention. Agents can lower some of these costs while competitors continually seek information whose value exceeds the remaining expense. Complete informational efficiency is a much stronger claim than cheaper routine analysis.

Our current product example establishes a workflow rather than an equilibrium result. DXAP's activity guide distinguishes evaluations, proposed actions, submitted orders and recorded fills. Those distinctions matter for adoption research: a population of active agents can perform many evaluations with limited executed demand. Counting agents or generated theses alone cannot establish a change in market efficiency.

Observe prices and crowding separately

We propose studying a predefined set of public releases and liquid instruments over successive calendar periods. Preserve original release availability times, hold the forecast horizons fixed at six hours, one day and three days, and record executable trading costs. Track the cost of reproducing a research workflow alongside the speed with which prices adjust after the release.

A supporting observation would be declining reproduction cost together with shrinking after-cost opportunities for the same public-signal family. Faster price adjustment would strengthen the mechanism. Interpret the comparison carefully: volatility, venue design and changes in the release itself can explain part of the shift. Agent adoption needs an independently measured exposure, with contemporaneous comparison markets where feasible.

Crowding has a different signature. Measure concentration in order direction and timing, implementation shortfall and behavior when many positions unwind. A market can incorporate public news quickly while suffering concentrated exit demand later. Better information processing and fragile common positioning can coexist. Aggregate forecasting accuracy would hide that combination.

The falsifiers are concrete. If reproduction becomes cheaper and measured adoption rises while the same cost-adjusted opportunities persist, the proposed erosion mechanism weakens. If price adjustment accelerates without relevant adoption, another mechanism deserves credit. If apparent gains disappear when release timestamps are corrected, the original opportunity was an evaluation artifact.

A stronger test would compare equally specified workflows under controlled adoption in a market simulation with explicit liquidity. That would identify a mechanism under the simulator's assumptions. Public-market observation would still be required to establish whether the effect transfers. We have proposed these tests; no completed DXRG equilibrium estimate is implied.

We expect broader capability to make routine interpretation less scarce. The resulting market can support more informed owners while offering smaller rewards for familiar public-signal trades. The valuable research question then becomes which remaining uncertainty is worth resolving at the new cost, and whether acting on it remains economical after everyone else can ask the same question.