Tldr:
- Dragonfly’s Qureshi compares today’s AI agents to a mouse in 1964, warning that adoption will take much longer than expected.
- OpenClaw is still buggy and unreliable for financial tasks as models run outside of their training distribution today.
- The x402 protocol only processes about $1 million per day, which confirms that the market is still in a patchwork phase.
- Qureshi expects a new generation of models to emerge within months, but says it will take several more years to reach the early majority.
Agent payments It is gaining traction as a discussion point across cryptocurrency and fintech circles globally. However, a prominent voice from one of the most well-known cryptocurrency investment firms is urging caution about timelines.
Haseeb Qureshi, managing partner at Dragonfly Capital, recently shared what he called his “more pessimistic stance” on the topic.
While he believes agents will eventually reshape how money moves, he believes the technology is still far from ready for mainstream use.
Qureshi Dragonfly points to history as a cautionary standard
Qureshi based his warning on a well-known piece of technology history. He pointed to the computer mouse, first invented in 1964, as a parallel to today’s artificial intelligence agents.
This invention clearly referred to universal personal computing, but it took many additional years for widespread adoption. His point is that discovering a transformative technology early does not mean it arrives on time.
OpenClaw is at the center of its current doubts about effective preparedness. The Dragonfly executive described the tool as buggy, complex, and unsuitable for managing real financial assets.
She regularly makes bad decisions and, as he puts it, “goes bankrupt doing stupid things.” These are not simple limits – they reflect a structural gap between agent capability and real-world task requirements.
The main problem, according to Qureshi, is that current models handle tasks well outside their training distribution. This mismatch results in the erroneous and unreliable behavior that users routinely encounter.
No major laboratory has directly applied reinforcement learning to it OpenClaw Interaction Effects However, these effects carry a strong training signal that laboratories have not yet exploited.
Once a tester trains purpose-built models on the agent’s task data, a significant improvement in performance is expected. Qureshi noted that every major AI lab is working on this, because the commercial prize is clearly visible.
The release of this model will likely arrive within months rather than years. However, even this significant event would only lead to the end of the patchwork era, not the beginning of mass adoption.
Live payment data supports Dragonfly partner’s cautious stance
Qureshi pointed to real protocol data to support his position on the current market situation. The x402 protocol processes approximately $1 million in daily volume at present.
The automated payment protocol records even smaller numbers. Together, these numbers confirm that the current user base consists almost entirely of early experimenters.
The Dragonfly CEO also relied on a widely cited framework from investor Chris Dixon. The idea is that what technically curious people do on weekends today, the wider public will be doing in ten years.
This pattern has been repeated consistently across major waves of technology, from the Internet to mobile. Agent payments appear to exist at the beginning of that same cycle.
Qureshi has plotted the full adoption curve to give context to what comes next. After the patching phase closes, the market enters the early adoption zone, which will take time to mature.
The early majority follows, then the late majority and late adopters eventually come. Each stage has its own timeline, and none of them collapse quickly.
for now, dragonfly The partner sees agents as a long-term story that the industry should not rush into. The direction of technology is clear, and the destination is not in doubt.
What remains uncertain is how long each stage of adoption will actually take. He argues that this uncertainty is exactly what cryptocurrencies used to downplay.
