Cursor vs Claude Code

The author explains Claude Code's design philosophy of terminal-first approach as a deliberate choice to lower adoption friction.

The author states features like pet mode and sub-agent displays are designed to look impressive in screenshots, and that Anthropic optimizes for 'the feeling of productivity.'

The author argues Anthropic's philosophy is 'if more tokens solve the problem, use more tokens,' citing sub-agent mode as an example that burns tokens aggressively.

The author cites artificial analysis benchmarks showing GPT-55 used half as many tokens to get a better score compared to Anthropic models.

The author states Anthropic wants users in their CLI using their features and prevents programmatic API access without charging more money.

The author says his agentic coding app 'would not exist if it wasn't for the Codex app server, which is part of the open source Codex CLI.'

The author describes attempting to integrate Cursor into T3 Code and finding the cursor CLI 'super behind' with hardcoded, outdated models.

The author claims the tool Anthropic employees use internally is not the same as the public version, leading to poorly tested features for end users.

The author states 'the Codex app that we're using... is the exact same thing that they're using at OpenAI,' citing conversations with friends at the company.

The author states 'They use the shit out of the things that they are doing. They obsessively test every detail before taking the exact thing they're using internally and throwing it at us.'

The author demonstrates login failures, lack of thread sync, and poor UX in the desktop app, concluding 'they straight up do not use the desktop app.'

The author states 'They didn't build this app for you to use it. They built this app so that it would exist if people stop asking for it.'

The author contrasts Codex's minimal UI with Claude Code's flashy animations, noting Codex 'isn't trying to feel like a slot machine. It's just trying to get work done.'

The author argues that computer use verification ends up being 'a lot less work than spinning up 15 agents to go triple check every line of code.'

The author states 'I still feel like 46 and especially 47 are more so regressions than they are improvements' and that 'Opus got worse twice.'

The author states '52 to 53 to 54 to 55 were massive improvements' and contrasts this with Anthropic where 'Opus got worse twice and OpenAI models got better three times.'

The author says 'Anthropic has to make up for this somehow... by doing a shitload of cool looking features and trying to make moments happen.'

The author describes Cursor's cloud sandbox as spinning up 'a full graphical interface Linux instance' that can run the full app and use computer use to test changes.

The author demonstrates that from Slack, someone can ask the cursor bot to 'spin up an agent to fix this' and it 'responds in the thread with a video of the fix.'

The author summarizes his recommendation: 'Cloud code for unmotivated devs or bad devs that want to feel like they're productive. Codex for skeptical devs that want to use these tools in a way that is productive.'

The author recommends Cursor cloud as 'one of the best solutions to set up for your team so nontechnical people can kick things off in Slack.'