Choosing TypeScript Matters More Than Ever

Keep choosing TypeScript — it matters more than ever.

LLM output is not "new machine code." You should learn a language well, and TS is the perfect choice.

Why should we care about the language?

Consider the world we have built for the agents: a tower of abstractions, 0/1 at the bottom, climb the tower —

ASM, compilers -> C, TS/Go/Python -> frameworks -> applications

We spent most of our time in the language layer in years past.

People who dabble in predictions — where will they land? Squash lower lower languages, including high-level languages.

The truth is in the middle: between irrelevancy and English being a programming language.

English is not a good programming language.

Paul Graham — "… writing in a language…"

We wouldn't be committing code, we would be committing prompts (and we're not, yet).

• WASP — framework

Why TypeScript?

Software engineering is compression — time to compress personal context.

types -> code -> docs -> memory

(overlapping nested concepts)

MEMORY is outsized in space.

Shoot first, ask questions later? Code is cheap? Right?

Imagine there is no TypeScript; what's reasonable as far as compressing complexity?

Did we answer our domain questions? NO! Where are the limits, and so on. So we didn't compress memory into types.

So, hey! Let's add documentation! We just moved some of our memory to docs.

Docs are not the harness. Docs are not deterministic, and there is no guarantee that docs match reality.

Plan B: Tests

We can move some memory into code with test coverage and case exhaustion, and add happy path.

Just use JS, not TS, and add a bunch of tests if you want to throw off 100k lines of code a day 🤑

• Reduce entropy by adding extra code.
• Tests expect us and the agent to generalize.

Types

Move context from memory and code to Types. Encode information into the type.

In the naive case, we still don't answer constraint-related questions.

If you know something, the types should know it as well!

What if we start with the questions? With the Types?

Fred Brooks: "Show me your flowcharts and conceal your tables, and I shall continue to be mystified. Show me your tables, and I won't usually need your flowcharts; they'll be obvious."

We could define what we want in a prompt, which is basically a specification.

We should instead take a step back and define our type MANUALLY. That's better than using English; it's more precise.

• We can compress our memory AND preserve context by reaching for the right type.

• Think about the questions related to our domain, and then define the appropriate types.

• Discover the domain WHILE building the model.

• GrillMe — skill to help refine domain modeling.

He thought about talking about "branded types", result types, ensuring exhaustiveness, smart constructors, edge type safety — decided to make it simpler.

Rust can do all of this, and other stuff; it may do stuff we don't need.

TypeScript continues to evolve

• modern and safer
• rewritten in Go
• better errors
• most popular language
• removed legacy baggage