Dragonfly’s Haseeb Qureshi Warns Agentic Payments Are Not Ready for Mass Adoption

TLDR:

• Dragonfly’s Qureshi compares today’s AI agents to the 1964 mouse, warning adoption will take far longer than expected.
  
• OpenClaw remains buggy and unreliable for financial tasks as models operate outside their training distribution today.
  
• The x402 protocol processes only around one million dollars daily, confirming the market is still in its tinkering phase.
  
• Qureshi expects a new model generation within months, but says reaching the early majority will still take several more years.

Agentic payments are gaining momentum as a talking point across crypto and fintech circles globally. Yet a senior voice from one of crypto’s most recognized investment firms is urging caution on timelines.

Haseeb Qureshi, a managing partner at Dragonfly Capital, recently shared what he called his “most bearish take” on the subject.

While he believes agents will eventually reshape how money moves, he argues the technology remains far from ready for mainstream use.

Dragonfly’s Qureshi Points to History as a Cautionary Benchmark

Qureshi grounded his warning in a well-known piece of technology history. He referenced the computer mouse, which was first invented in 1964, as a parallel to today’s AI agents.

That invention clearly pointed toward mass personal computing, yet widespread adoption took many additional years. His point is that spotting a transformative technology early does not mean it arrives on schedule.

OpenClaw sits at the center of his current skepticism about agentic readiness. The Dragonfly executive described the tool as buggy, complicated, and unfit for managing real financial assets.

It regularly makes poor decisions and, in his words, “goes bankrupt doing stupid shit.” These are not minor rough edges — they reflect a structural gap between agent capability and real-world task demands.

The core problem, according to Qureshi, is that current models are handling tasks well outside their training distribution. That mismatch produces the erratic and unreliable behavior users routinely encounter.

No major lab has yet applied reinforcement learning directly to OpenClaw interaction traces. However, those traces carry strong training signal that labs have not yet tapped.

Once a lab trains purpose-built models on agentic task data, a major performance improvement is expected. Every major AI laboratory is working toward this, Qureshi noted, because the commercial prize is clearly visible.

That model release will likely arrive within months, not years. Still, even that milestone will only mark the close of the tinkering era, not the start of mass adoption.

Live Payment Data Backs the Dragonfly Partner’s Cautious Stance

Qureshi pointed to real protocol data to support his position on where the market currently stands. The x402 protocol is processing roughly one million dollars in daily volume at present.

The Machine Payment Protocol is recording even smaller figures than that. Together, those numbers confirm the current user base consists almost entirely of early experimenters.

The Dragonfly executive also drew on a widely cited framing from investor Chris Dixon. The idea is that what technically curious people do on weekends today, the broader public will be doing within ten years.

That pattern has played out consistently across major technology waves, from the internet to mobile. Agentic payments appear to be sitting at the very beginning of that same cycle.

Qureshi mapped out the full adoption curve to give context to what comes next. After the tinkering phase closes, the market enters early adopter territory, which itself will take time to mature.

The early majority follows that, and then comes the late majority and eventual late adopters. Each phase carries its own timeline, and none of them collapse quickly.

For now, the Dragonfly partner sees agents as a long-term story that the industry should not rush. The technology direction is clear, and the destination is not in question.

What remains uncertain is how long each phase of adoption will actually take. That uncertainty, he argues, is precisely what crypto has a habit of underestimating.