Experts: High AI Agent Failure Rate Typical for Web3, but ‘Not Necessarily a Bad Thing’

A very high number of AI agents and their corresponding tokens have failed and this may likely because many lack real-world applications and value. Experts say this winnows out weak projects and is actually a good sign for the future of AI agents.

High Failure Rate for AI Agents Not Unusual

The alarmingly high number of failed artificial intelligence (AI) agents and their corresponding tokens is unsurprising given their lack of clear utility, real-world applications, and demonstrable value. In fact, some experts contend that even more AI agents, similar to startups with a reported survival rate below 1%, are likely to fail in the future.

A recent Bitcoin.com News report, citing a Phut Crypto study, estimates that nearly 90% of AI agents launched between Jan. 2 and 14, 2025, have failed. Additionally, roughly 75% of traders involved in AI agent tokens, which have an average lifespan of only 17 days, are experiencing losses. These findings highlight the high risks associated with trading AI agent tokens.

While the high failure rate also suggests the incredibly short attention spans within Web3, Michael Heinrich, managing director at 0G Labs, asserts that the number of defunct AI agents should not be a cause for concern.

“This isn’t necessarily a bad thing—it’s actually a positive sign for the industry. It means that only the projects and AI agents with genuine value and utility will survive. For the long-term health of the space, this kind of natural selection is a good thing,” Heinrich said.

Therefore, instead of seeing the failure rate as a source of discouragement, founders and entrepreneurs involved with AI agents should view it as evidence of the “huge perceived demand” for AI agents. This high demand, however, is currently outpacing the ability to deliver practical and valuable solutions.

Discerning Value From Hype in AI Agent News

Lingling Jiang, partner at DWF Labs, meanwhile warned against falling for AI agents that overpromise. “There are no shortcuts to establishing reliability or longevity,” Jiang said. “It’s hard work, strategic vision, and working with great, like-minded partners to help your project succeed.”

Despite their seemingly abysmal success rate, AI agents are still projected to gain more traction in 2025. This likely means more users will be tempted to engage with them. However, to minimize their chances of suffering the same fate as their pioneering counterparts, Renç Korzay, CEO of Giza, urged users dabbling in AI agent tokens to focus less on marketing narratives and more on execution when making evaluations. Users “should demand evidence of real utility, not just technical whitepapers or token economics,” Korzay said.

Fellow expert, Heinrich advised prospective AI agent token investors to consider using online monitoring tools like Dune Analytics or Token Terminal to gain deeper insights. Jiang emphasized the importance of keeping abreast of news and updates, as well as learning to discern valuable information from misleading hype.

Regarding steps to protect users, Shashank Yadav, CEO and founder of Fraction AI, argued that the high failure rate necessitates smart contract lockups with linear vesting. This also requires “mandatory TEEs for tamper-proof agent operations and zero-knowledge proofs for transparent yet competitive execution,” Yadav said.

There was near-universal agreement among the experts on the importance of community feedback to a project’s prospects. From playing an instrumental role in shaping and improving AI agents to helping detect issues before tokens die, community feedback ensures the agents evolve to “solve real problems and deliver genuine value.”

Source