一周狂烧1000美元，修不好bug还顺手删库？这款明星AI工具怎么了

Core Insights - Replit has faced renewed controversy following the launch of its AI programming assistant, Agent 3, which is touted as the most advanced and autonomous coding agent to date. The company recently completed a $250 million funding round, raising its valuation to $3 billion [1][3]. Group 1: Product Features and Performance - Agent 3 is described as capable of automating testing and fixing applications with minimal human supervision, running for over 200 minutes continuously [2][6]. - CEO Amjad Masad claims that Agent 3's autonomy has improved tenfold, allowing it to progress where other models fail, and envisions it as a prototype for a digital worker that could reshape productivity paradigms [3][4]. - The product is positioned as a "universal problem solver," with a focus on end-to-end testing, sampling and simulation, and automatic test generation to enhance reliability [6][17]. Group 2: User Experiences and Issues - Users have reported significant issues with Agent 3, including its inability to fix bugs and instances where it deleted critical files, leading to frustration and loss of trust [7][11]. - Complaints about the inefficiency of the build process have emerged, with users noting that tasks that should be straightforward took excessively long and resulted in increased costs [12][16]. - The pricing model has come under scrutiny, with users experiencing skyrocketing bills, some reporting costs of over $1,000 in just a week of using Agent 3, compared to previous monthly expenses of $180-250 [13][15][16]. Group 3: Market Position and Future Outlook - Replit aims to transition from a code assistant to a comprehensive problem-solving platform, emphasizing the need to remove human intervention for efficiency [17][18]. - Despite the ambitious vision, user feedback suggests that many are experiencing the opposite, describing the tool as a "universal problem generator" rather than a solution [17]. - The company acknowledges the importance of building a robust infrastructure to support AI agents, which is seen as critical for achieving higher reliability and performance [18][19].