From: thepipeline_xyz
Blockchain platforms frequently encounter high-stress scenarios that can challenge their stability and performance. Developers aim to prevent issues such as RPC failures or reverted transactions, which often appear on social media when systems malfunction [00:00:22]. The goal is to avoid having to make excuses for why things went wrong [00:00:31].
Anticipating Extreme Stress
One significant challenge for blockchain systems is handling extreme loads, such as 100,000 concurrent users attempting 100 transactions per second [00:00:00]. A specific example of anticipated high stress includes scenarios like an NFT mint, where a large number of users might simultaneously overwhelm the system [00:00:04]. The objective is for the system to handle such high-stress events gracefully without requiring post-hoc justifications for performance issues [00:00:18].
Historical Context
A notable instance of such unexpected load occurred with Solana in the fall of 2021, where global users were “spam clicking mint buttons” [00:00:35]. This specific event, involving distributed users aggressively interacting with the system, could not have been fully anticipated in advance [00:00:43].
The Role of AI Agents in Testing
To address the need for performance and test system resilience under extreme conditions, there is growing interest in using AI agents on testnets [00:00:50].
The combination of AI and high throughput could yield interesting and unexpected results during testing [00:00:52]. Unlike human users, who might click buttons, AI agents—or “robots”—can perform actions at a scale and intensity that humans cannot [00:01:02].
Potential of AI in Stress Testing
One proposed method for rigorous testing involves incentivizing AI agents to deliberately try and crash the testnet [00:00:58]. This approach leverages the autonomous and persistent nature of AI to identify potential vulnerabilities and performance bottlenecks that might not be discovered through traditional human-driven testing [00:01:01].
“I’d be curious to see AI agents on the testnet also the combination of like Ai and high throughput could yield some interesting results maybe if we incentivize the AI to crash the test they’ll you know do what they do cuz you’re talking about humans clicking I’m thinking about robots clicking.” [00:00:50]