Ethereum’s co-founder Vitalik Buterin spoke out against the emphasis on making AI models more agentic. Buterin believes open-weight models with more human input are the way forward.
Vitalik Buterin spoke out against the drive to make AI more agentic instead of making it more capable of getting more feedback from humans.
Echoing something @karpathy recently said, it does frustrate me how a lot of AI development is trying to be as "agentic" as possible, when actually creating *more* paths for human input both creates a better output (now for quite a while going forward) and is better for safety.
— vitalik.eth (@VitalikButerin) August 11, 2025
Agentic AI, creating a notable personality, has been a success with end users as a novelty feature. In crypto, agents with a clear brand and personality became the norm, for their potential to drive engagement on social media. As Cryptopolitan reported earlier, Vitalik also saw limited utility in using AI to enhance crypto activity.
Buterin replied to a previous thread by Andrej Karpathy, where he explained that the most widely used AI models spend too much resources on offering agentic solutions without explicit prompts.
I'm noticing that due to (I think?) a lot of benchmarkmaxxing on long horizon tasks, LLMs are becoming a little too agentic by default, a little beyond my average use case.
For example in coding, the models now tend to reason for a fairly long time, they have an inclination to…
— Andrej Karpathy (@karpathy) August 9, 2025
Agentic AI has promised to scale tasks at a speed and efficiency impossible for humans. However, the human feedback and input approach may bring more accurate results for specific queries and tasks.
Buterin focused more on output and the ability of humans to receive the desired output. AI agents, on the other hand, have limited behaviors that are mostly used as marketing ploys. Recently, an AI agent was tasked with selecting an NFT as investment, and ended up overpaying 89 ETH for a low-rarity item.
Vitalik Buterin: humans are the steering wheel of AI
Previously, Buterin detailed some of his ideas about AI, considering AI as the engine and humans as the steering wheel.
Buterin also proposed the avoidance of a single AI personality, instead relying on human input, incentives, and swarms or collaborations of AI agents or LLM to achieve the desired results. In an essay from February, Buterin advocated for human judgment, either in individual form or as a jury or a collection of DAO voters.
The AI-human interaction also takes into account edge cases involving query resolutions and yes/no outcomes. In crypto, those outcomes and outputs may be key to prediction markets. Coordination and decisions are also one of the conflict points in decentralized systems, as Buterin noted.
AI agent space attempts a recovery
AI agent tokens are still a relatively strong narrative, recovering in the past few weeks. As a whole, the AI agent token space regained its market capitalization, reaching $6.45B.
Ethereum AI agents are in the green for the past week, reflecting interest in market analysis, DeFi activity, and trading. Agents themselves may not be involved in trades, but they are battling for mindshare on social media.
Ethereum remains one of the top chains for AI agents, with 27% of mindshare. The L2, Base, carries 30% of the AI mindshare. Solana still carries 19% of AI agent mindshare.
Overall, legacy agents with first-mover advantage are more resilient, mostly due to the cult communities around them. Demand for new agents has slowed down, and the gains come from revisiting some of the successful older projects.
Cryptopolitan Academy: Tired of market swings? Learn how DeFi can help you build steady passive income. Register Now
