<h2><strong>TL;DR</strong></h2>
<ul>
<li>**Cursor is a VS Code fork **processing 400M+ AI requests per day, built by ~100 engineers, reaching $1B ARR in 24 months, faster than any B2B SaaS company in history .<a href="https://theaiengineer.substack.com/p/how-cursor-actually-works#footnote-1-188185638" target="_blank" rel="noopener nofollow">1</a></li>
<li>**The core innovation is speculative edits: **your existing source code serves as “draft tokens,” pushing a 70B model to ~1,000 tokens/second (13x speedup) because most of an edited file is identical to the original.</li>
<li>A <strong>reinforcement learning loop</strong> retrains the Tab model every 90 minutes based on what you accept and reject, deploying new checkpoints multiple times daily at 400M+ requests.</li>
<li><strong>The transparency gap is real.</strong> No SWE-bench scores. A removed blog post. Proprietary benchmarks. The engineering is impressive, but the evidence is harder to verify than it should be.<img src="https://media.daily.dev/image/upload/s--fvYMXpLF--/f_auto/v1772205017/ugc/content_07b9098a-cd80-4dab-9e15-6fd46f7b53e0?_a=BAMAMiiu0" alt="795b0a18-361e-43a0-9a4a-31302999fcd0_6569x4220.webp"></li>
</ul>


Paolo Perrone

Cursor, a VS Code fork built by ~100 engineers, has reached $1B ARR in 24 months by processing 400M+ AI requests per day. Its core technical innovation is speculative edits: using existing source code as draft tokens to push a 70B model to ~1,000 tokens/second (a 13x speedup), since most of an edited file remains unchanged. A reinforcement learning loop retrains the Tab model every 90 minutes based on user accept/reject signals, deploying new checkpoints multiple times daily. However, the post also flags a transparency gap: no SWE-bench scores, a removed blog post, and reliance on proprietary benchmarks make the engineering claims difficult to independently verify.

https://theaiengineer.substack.com/p/how-cursor-actually-works

<p>Don’t put links in the title of the article…</p>
