After Code: How Cursor Is Redefining Software Development
Michael Truell
Co-founder & CEO, Anysphere (Cursor)
2024
The Vision
After Code: The Next 10 Years
"Being an engineer will start to feel like being a logic designer. It will be about specifying your intent for how exactly you want everything to work—less about the how, more about the what."
Move away from formal programming languages (TypeScript, Rust, C)
Evolve toward pseudocode and English-like logic representation
Keep humans in the driver's seat with complete control
Enable fast iteration loops, not weeks of waiting
The Path
From Boring to Ambitious: How Cursor Found Its Moment
Started wrong: Spent 4 months on mechanical engineering tools
Realized: Weren't passionate about the space, didn't have domain expertise
Pivoted right: Noticed existing AI coding tools lacked ambition
The lesson: Don't chase boring industries if the hot space has obvious flaws
The real opportunityEvery major category has a high ceiling when AI is involved. Look for gaps in ambition, not gaps in competition.
$0
starting point
$100M
ARR in 20 months
$300M
ARR at 2 years
Inflection point
Growth has been consistent exponential. Early, the numbers feel slow. By year 2, it's undeniable.
The Architecture
The Stack: Ensemble of Custom Models
Expected: Use only large foundation models (ChatGPT, Claude, Gemini)
Reality: Every magic moment in Cursor involves a custom model
The trick: Don't reinvent the wheel—focus on foundation model weaknesses
The wins: Massive speed gains, cost efficiency, specialized use cases
Autocomplete: The 300ms Model
Code is predictable. When you edit one part, you know exactly what else needs changing. We built models that predict across multiple files in under 300ms.
Search + Fill Architecture
Small models search the codebase to find context for big models. Big models think high-level. Small models fill in the detailed diffs at speed.
Flavor Change
Taste > Carefulness
As AI gets better, being careful (finding every bug) matters less
Taste—having the right vision—matters more
Taste applies to logic AND visuals, not just design
The future engineer: logic designer with strong taste, less debugging perfectionist
The skill shiftEngineers will spend less time catching mistakes and more time envisioning what should be built. Taste isn't vague—it's having the right idea.
Contrarian Truths
What Everyone Gets Wrong About Cursor & AI Code
✗Coding will stay text editing foreverINSTEAD →✓ In 10 years, pseudocode and intent-driven UI replace traditional code entirely.
✗Just use big foundation modelsINSTEAD →✓ Custom models for specialized tasks unlock speed, cost, and quality at scale. Ensemble beats monolith.
✗Moats don't exist in AIINSTEAD →✓ Moats are fragile but real—they come from relentless product ambition and continuous leapfrogging, like search in 1999.
✗You either hire fast or you failINSTEAD →✓ We hired too slow early on—and it worked out fine. Hiring slowly with intention beats hiring fast and iterating.