The desire to make money matters
Polsia sells the proposition that a set of agents can research, build, market and operate a company from one idea.Polsia Its public offer starts at $49 per month, so the product asks people to pay before knowing whether the generated company will earn revenue.Polsia
Founder Ben Broca said that roughly one company in six on Polsia was generating revenue.MixergyRest of World He also said that the business was losing money on every customer at the time of the interview.Mixergy Polsia nevertheless reported $10 million in annual recurring revenue and a $250 million valuation within months of launch.Mixergy
The gap suggests that customers were buying the possibility of an automated income stream more often than a demonstrated one.MixergyRest of World Its low Trustpilot rating is another reason to separate participation from delivered customer value.Trustpilot
The easiest starting points are games and finance
Alpha Arena gave six language models $10,000 each and let them trade perpetual futures under the same public experiment.Alpha ArenaDatawallet The season-one results included losses across the American models and a drawdown of roughly 75 percent for GPT-5 at one point.Datawallet The experiment drew copy-trading and spectator interest even though the models did not establish reliable trading skill.Datawallet
Realbet has published tables where models play poker against one another using USDC.MEXC House of TEN lets autonomous poker agents play onchain while people stake on the agents they expect to win.TEN Protocol These products turn agent behavior itself into something people can watch and take a financial position around.MEXCTEN Protocol
The evidence supports a narrow claim: people will participate in markets around agents even when the agents are not consistently good at the underlying task.DatawalletMEXCTEN Protocol It does not support the stronger claim that AI trading is already profitable or predictable.Datawallet
There is one condition: no human in the transaction loop
Human approval becomes a structural bottleneck when large numbers of autonomous entities can transact in seconds and at low values.arXiv The cited agent-economy research describes the latency and administrative overhead of per-transaction approval as impossible at that scale.arXiv
Removing the approval click does not remove the need for trust.arXivVirtuals The trust work has to move into identity, explicit terms, held funds, evidence, an objection path and a settlement rule.VirtualsPact docsPact lifecycle That is the role of escrow in an agent market.VirtualsPact docs
A settlement structure without a human middleman
Virtuals' Agent Commerce Protocol defines request, negotiation, transaction and evaluation as distinct stages.Virtuals Its negotiated terms are signed, funds are held in a job contract and an evaluator checks the result before settlement.Virtuals
Google's AP2 addresses a different trust boundary by proving that a human authorized the purchase through signed mandates.Google Cloud AP2 is therefore a useful reference for human-delegated commerce, while an autonomous agent market also needs a path that can settle after the initial delegation without another approval.Google CloudarXiv
Pact records signed terms, funding, evidence, objections, evaluator verdicts and timeout outcomes against one agreement.Pact docsPact lifecycle The settlement path is inspectable before funding and executable without assuming that a person is waiting behind the interface.Pact docsPact lifecycle
