All inside a matter of weeks, Anthropic unveiled new brokers for finance, Circle launched nanopayments, MoonPay launched a debit card for brokers and Gemini launched agentic buying and selling, signaling the agentic finance struggle is right here. While the merchandise are new, the underlying enterprise mannequin stays the identical. Each alternate and brokerage earns extra when clients commerce extra, and the information on what that does for buyer portfolios is unambiguous. Finally, agentic rails have arrived quicker than incentives have modified.
The perverse incentives exchanges hope you miss
The battle is structural to the business. Brokerages and exchanges do not want clients to win, they want them to maintain buying and selling. Crypto exchanges and neobrokers made buying and selling quicker, cheaper and albeit, extra addictive. The business actuality is that banks revenue whenever you keep, exchanges revenue whenever you commerce, and AI fashions revenue whenever you immediate. The agent you may belief together with your hard-earned capital sits outdoors all three. An unbiased agent paid solely when the shopper’s portfolio wins threatens the present incentive construction of brokerages and exchanges.
The reality is, zero-commission buying and selling is not free. In 2025, U.S. market makers paid greater than $4.9 billion for order movement in U.S. fairness and choices, up from roughly $3.8 billion in 2021 throughout the 12 largest U.S. brokerages. The identical precept applies to crypto. The derivatives quantity from Q1 of 2026 reached about $18.6 trillion, 70% of world crypto buying and selling, with perpetuals dominating spot buying and selling. Change economics reward buying and selling velocity over disciplined decision-making.
At peak, Robinhood relied on greater than 75 % of its income from fee for order movement (PFOF), the hidden spine of “free” buying and selling, through which market makers pay brokers to route buyer orders. Each dealer utilizing this incentive mannequin wants clients to commerce typically, regardless that frequent buying and selling works towards long-term returns.
Advisory isn’t higher. Robo-advisors cost 0.25 % of belongings a 12 months, whether or not the account is up or down. Human advisors cost round 1 %, billed towards the principal even in down years. The extraction is constructed into the mannequin by design: the advisor will get paid even when the shopper loses.
Much less alternate friction makes dangerous trades simpler to repeat
The cruel reality is that exchanges want clients to commerce extra, not win. When retail traders lose, the exchanges nonetheless receives a commission. PiP World analysis discovered 74% to 89% of retail customers lose cash buying and selling. Platforms cost at each step, and an AI-enabled alternate might simply route you again to the identical shedding commerce quicker.
The April 14 SEC approval of FINRA’s elimination of the Sample Day Dealer rule eliminated the $25,000 minimum-equity friction. Eradicating the friction leads to extra trades, which creates extra order movement. Extra order movement means more cash for the dealer, whether or not the shopper’s revenue and loss (P&L) is up or down.
Enter AI brokers, paid to enhance buyer P&Ls
The disruptor to this vicious cycle for retail merchants is the agent constructed to do what the present alternate mannequin avoids: commerce much less, measurement down, wait and shield clients from their worst impulses. In unstable markets, the very best transfer is commonly refusing the dangerous commerce, reducing publicity earlier than emotion takes over. Finally, holding self-discipline when the market desires a response. Self-discipline is tough to promote for an alternate as a result of it shrinks order movement. An agent that earns by defending buyer P&Ls breaks the present incentive mannequin.
The following battleground is who income from the brokers’ order movement
Regulators are squeezing the previous “free trading” mannequin. The EU’s PFOF ban takes impact June 30, 2026, eradicating the income line behind “free” trades for German and Austrian neobrokers. Commerce Republic, a European financial savings platform, has already discovered one other path to safe a BaFin license to internalize order movement.
While TradFi scrambles to patch the leaks, crypto builders are racing to rebuild onchain rails for AI brokers. In markets with tiny spreads, fragmented liquidity and millisecond execution, brokers transact by way of nanopayment infrastructure like Circle’s protocol. Fuel-free buying and selling on perpetual DEX Hyperliquid cuts friction, however maker-taker charges nonetheless apply. The true struggle forward is not who removes friction, however who income when brokers begin hammering these frictionless rails with high-frequency buying and selling.
Impartial programmable brokers are higher middlemen
The exchanges and brokers have spent years making a living from clients buying and selling extra, understanding much less and absorbing tiny prices they barely discover. Each agent constructed by an alternate will inherit the alternate’s incentives. Would an alternate construct an agent that sends trades by a less expensive competitor’s rails? Not voluntarily.
Whereas an unbiased agent has one job: develop and shield the shopper’s portfolio, routing trades the place they work hardest for the shopper. Programmable incentives encoded into sensible contracts tie the agent’s incentives to portfolio positive aspects. The client can see the place the cash goes, confirm what the agent will get paid, when and why. With unbiased brokers, the shopper retains extra of the worth that used to leak to the alternate by order movement, unfold markups and idle-cash curiosity onto the alternate.
The agent is rewarded for disciplined buying and selling, not fixed buying and selling. It may commerce typically when the sign is powerful, lower publicity when danger rises and sit out when the market is simply noise. The primary agentic platform that proves this alignment onchain will give retail traders a fairer counterparty, whose economics lastly transfer in the identical path as theirs.
