Back to Mo Bitar

Cursor wants to KILL programming

Mo BitarJune 17, 202613m
In a Nutshell

SpaceX is acquiring Cursor, whose CEO argues programming will shift from writing formal code to describing software intent in structured English-like pseudo-code. Meanwhile, cheap Chinese open source models like GLM 5.2 are undercutting subsidized Western APIs on cost and performance, threatening the current pricing model. AI is already eroding non-fiction book sales by replacing lookup-style information with instant, personalized chatbot responses.

AI-Generated Notes

These notes were generated by AI and may contain inaccuracies.

It is June 17th, Wednesday. A letter from Secretary of Commerce Howard Lutnik informs Dario Amodei that a license is required for Claude Mythos 5. The letter instructs that license applications must be submitted through SNAPR at snapr.biz.gov, with specific notation in the additional information box referencing the letter and attaching a copy. Dario must navigate the same bureaucratic website process as everyone else.

Anthropic's left-leaning positioning creates friction with the current administration. Dario speaks in technical terminology that creates a language barrier with regulators.

SpaceX has agreed to acquire Cursor in an all-stock transaction. SpaceX currently holds the fifth or sixth most valuable position among US companies with a $2.5 trillion valuation. Nvidia generates $215 billion in revenue with $117 billion in profit.

Cursor announced the release of its own homegrown model with 1.5 trillion parameters, matching the size of Claude Opus and GPT 5.5. The model was pre-trained on 100k GPUs. Cursor grew from $100 million to $1 billion to $2 billion ARR in 18 months, setting the record for fastest SaaS to reach $2 billion. The company has over 1 million paying users, with 60% enterprise customers and 67% of the Fortune 500 represented.

Cursor previously accounted for nearly 50% of Anthropic's enterprise revenue in the early days. Elon Musk's acquisition priorities likely include reducing Cursor's dependency on Anthropic models.

Cursor CEO Michael Truel stated their goal is to create a new type of programming focused on describing intent to computers concisely. This approach emphasizes defining how software should work and look rather than traditional text editing in formal languages like TypeScript, Go, C, or Rust.

Truel identified problems with both existing visions: chatbot-style interfaces lack precision, while traditional code-based development will evolve. He envisions a future representation of software logic resembling English in document form, evolving programming languages toward pseudo-code that is tiered, easier to understand, and navigable at a high level.

The attempt to create this pseudo-code abstraction repeatedly results in recreating code itself, because software requires precision and predictability. Departing from code introduces randomness and non-determinism.

GLM 5.2 benchmarks appear exceptional, outperforming Opus 4.8 and GPT 5.5 on front-end coding if Claude Fable did not exist. The open source model costs 82% less than Opus 4.8.

$3,000 purchases 3.45 billion output tokens using Deep Seek V4 Pro, 682 million tokens using GLM 5.2, and 60 million tokens using Claude Fable 5. China is effectively addressing AI cost concerns through these economical models.

Microsoft is exploring DeepSeek over Anthropic for Copilot as the service moves to usage-based pricing. The $200 plan on Anthropic delivers $8,000 of usage while OpenAI's plan delivers $14,000, functioning as highly subsidized offerings to build addiction and workflows.

Satia published that infrastructure must be built around models to make them replaceable, avoiding putting everything in Anthropic's hands. Without open source alternatives, subsidization wars between SpaceX, OpenAI, and Anthropic would continue, with the most capitalized player winning.

GLM 5.1 showed impressive benchmarks but did not permeate developer culture due to subsidized alternatives. True pricing discovery awaits the end of subsidies.

Tim Ferriss reported that AI killed sales of his non-fiction books. By 2026, sales are down 57% versus 2025. If trends continue, his catalog will sell 80% fewer print copies in 2026 than in 2022. The 4-Hour Body represents a lookup table where books were previously the best interface, but chatbots now provide faster, cheaper, and more personalized advice.

Ferriss predicts how-to YouTube videos will face disruption, as AI can watch 24-minute videos to extract needed 40-second segments. Content surviving AI disruption likely includes experiences beyond pure information: comedy, entertainment, storytelling, and fiction.

Keep Mo Bitar in your library

Save the videos and channels worth coming back to, and find them again in one place.