Apa Itu Von

According to Beating AI news, Shopify CEO Tobi Lütke is considering banning Claude Code within the company. It's not that Claude is ineffective, but rather that it insists on using its own CLAUDE.md. Shopify utilizes multiple agents, including Claude Code and Codex, while other tools can share AGENTS.md. However, Claude Code only reads CLAUDE.md. The two sets of documents will recursively layer in large code repositories, and if one layer fails to sync, someone will end up with incorrect rules. Shopify has already implemented automated checks to cover this issue, but Tobi feels this is unnecessary trouble. His request is quite simple: Claude Code should also read AGENTS.md, and maintaining a single version would suffice. Claude Code team member Thariq responded that they would make AGENTS.md easier to use. However, he then explained why Claude needs its own prompts, skills, and CLAUDE.md: the models differ, and specialized optimization yields better results. The short-term solution remains to reference AGENTS.md within CLAUDE.md, which does not address Tobi's concern. While it's acceptable to continue optimizing for Claude, why must project rules maintain a separate entry for Claude? Grok reviewed the previously leaked Claude Code source code and concluded more directly: the filename CLAUDE.md is hardcoded in the loading logic, and simply switching it to AGENTS.md would allow the original recursive discovery, loading, and referencing mechanisms to function without significant architectural changes. In short, AGENTS.md is usable; Claude Code just insists on keeping its own CLAUDE.md. Comments largely support Tobi, with DHH stating the distinction is unnecessary, Jesse Pollak expressing frustration, and Chamath Palihapitiya believing that maintaining two systems will inevitably lead to 'brain splitting.' Kun Chen suggested that the correct approach should be to default to reading AGENTS.md, and if Claude-specific optimization is truly needed, then CLAUDE.md should remain an optional choice, rather than the other way around.

According to Beating AI news, Shopify CEO Tobi Lütke is considering banning Claude Code within the company. It's not that Claude is ineffective, but it insists on using its own CLAUDE.md. Shopify employs various agents, including Claude Code and Codex. Other tools can share AGENTS.md, but Claude Code only reads CLAUDE.md. The two sets of documents will recursively layer in large code repositories. If one layer fails to sync, someone ends up with incorrect rules. Shopify has already implemented automated checks to address this issue, but Tobi feels this is unnecessary trouble. His request is quite simple: Claude Code should also read AGENTS.md, and maintaining a single version would suffice. Thariq, a member of the Claude Code team, responded that they would make AGENTS.md easier to use. However, he then explained why Claude needs its own prompts, skills, and CLAUDE.md: the models are different, and specialized optimization yields better results. The short-term solution remains to reference AGENTS.md within CLAUDE.md, which does not resolve Tobi's issue. You can continue to optimize for Claude, but why must project rules maintain a separate entry for Claude? Grok reviewed the previously leaked Claude Code source code and concluded that the filename CLAUDE.md is hard-coded in the loading logic. Simply changing it to AGENTS.md would allow the original recursive discovery, loading, and referencing mechanisms to function without major architectural changes. In short, AGENTS.md is usable; Claude Code just insists on keeping its own CLAUDE.md. The comments largely support Tobi. DHH stated that the distinction is entirely unnecessary, Jesse Pollak expressed frustration, and Chamath Palihapitiya believes that continuing with two systems will inevitably lead to 'brain splitting.' Kun Chen suggested that the correct approach should be to default to reading AGENTS.md, and if specific Claude optimizations are needed, then keep CLAUDE.md as an optional choice, rather than the other way around.

On August 26, according to Whale Alert monitoring, at 06:41 and 13:12 Beijing time today, USDC Treasury minted a total of 250 million USDC on the Solana chain in two transactions, amounting to a total of $500 million.