Dragonfly’s Qureshi Warns Agentic Payments Are Years Away
Dragonfly managing partner Haseeb Qureshi says the crypto market is moving too fast on agentic payments. He argues real-world adoption is still years away because today’s agentic payments tools are unreliable, overly complex, and handle real-money decisions poorly.
Citing projects like OpenClaw, Qureshi says models can fail outside their training data, creating bug risk and weak financial judgment in live payment flows. He also points to a key research gap: there is not yet reinforcement-learning training data from real agent interactions inside live payment systems.
Qureshi expects next-generation models in the coming months could improve performance, but mainstream usage likely remains delayed. To support the “experimental” stage, he notes daily volume is low—x402 processes around $1M/day, while MPP is even lower.
For traders, this is a reality check on hype cycles around agentic payments automation: near-term enthusiasm may fade, while the long-term narrative depends on whether reliability improves.
Bearish
Qureshi’s stance is explicitly skeptical: agentic payments may be a “huge trend,” but he frames it as firmly experimental today due to reliability issues and very low real-world transaction volumes. That framing can reduce near-term speculative enthusiasm and temper “payments automation” expectations, which can weigh on sentiment for related AI/automation narratives.
In the short term, traders often react to perceived timeline delays by rotating out of momentum trades tied to hype cycles. In the longer term, there is an offsetting positive: Qureshi still expects next-generation models within months, which could improve robustness once reinforcement-learning-on-real-interaction data catches up. However, until reliability and adoption volumes increase meaningfully, the dominant market effect is likely negative on sentiment rather than a direct positive catalyst.