Prompts: A Layer Above Declarative Programming?

Yuji Kuroki — Fri, 04 Jul 2025 12:41:20 +0000

Let’s chew on that question and follow the chalk-lines it draws to see what we uncover.

Let's kick off with two fundamental styles of programming: Declarative and Imperative.

Expression Style
Declarative
Imperative

When you ask for examples, it shakes out like this:

Expression Style	Language Type	Typical Medium (there’s always a spectrum)
Declarative	DSLs, programming languages	SQL, Prolog, Terraform
Imperative	Machine code, programming languages	Machine code, Assembly, C

The two styles map nicely onto What and How.

Layer	Expression Style	Language Type	Typical Medium (there’s always a spectrum)
What	Declarative	DSLs, programming languages	SQL, Prolog, Terraform
How	Imperative	Machine code, programming languages	Machine code, Assembly, C

In practice, the What layer often sprinkles in a dash of How. Here’s that nuance:

Layer	Writable Scope	Expression Style	Language Type	Typical Medium (there’s always a spectrum)
What	What, How	Declarative	DSLs, programming languages	SQL, Prolog, Terraform
How	How	Imperative	Machine code, programming languages	Machine code, Assembly, C

If there’s a What and a How, surely a Why sits on top, no?

Layer	Writable Scope	Expression Style	Language Type	Typical Medium (there’s always a spectrum)
Why	?	?	?	?
What	What, How	Declarative	DSLs, programming languages	SQL, Prolog, Terraform
How	How	Imperative	Machine code, programming languages	Machine code, Assembly, C

Zoom out one more notch and you can label what each layer actually describes:

Layer	Describes	Writable Scope	Expression Style	Language Type	Typical Medium (there’s always a spectrum)
Why	Outcome	?	?	?	?
What	Output	What, How	Declarative	DSLs, programming languages	SQL, Prolog, Terraform
How	Procedure	How	Imperative	Machine code, programming languages	Machine code, Assembly, C

Now, in modern software development we do have a tool for writing Outcomes—just hand them to an AI. Tell the model you want to boost your store’s conversion rate and it happily decomposes that wish into Change the CTA color, Resize the button, and so on. The “Expression Style” for this tier? Optative feels about right.

Layer	Describes	Writable Scope	Expression Style	Language Type	Typical Medium (there’s always a spectrum)
Why	Outcome	Why, What, How	Optative	Natural language	Prompts
What	Output	What, How	Declarative	DSLs, programming languages	SQL, Prolog, Terraform
How	Procedure	How	Imperative	Machine code, programming languages	Machine code, Assembly, C

Notice the pattern: the What layer punts the How to a runtime, and the Why layer punts the What to an agent.

Layer	Delegates Downward	Describes	Writable Scope	Expression Style	Language Type	Typical Medium (there’s always a spectrum)
Why	Leaves the What to the agent	Outcome	Why, What, How	Optative	Natural language	Prompts
What	Leaves the How to the runtime	Output	What, How	Declarative	DSLs, programming languages	SQL, Prolog, Terraform
How	Leaves execution to hardware	Procedure	How	Imperative	Machine code, programming languages	Machine code, Assembly, C

Push the thought experiment further and another layer emerges—one that generates the Why itself.

Layer	Delegates Downward	Describes	Writable Scope	Expression Style	Language Type	Typical Medium (there’s always a spectrum)
?	Leaves the Why to ?	?	?	?	?	?
Why	Leaves the What to the agent	Outcome	Why, What, How	Optative	Natural language	Prompts
What	Leaves the How to the runtime	Output	What, How	Declarative	DSLs, programming languages	SQL, Prolog, Terraform
How	Leaves execution to hardware	Procedure	How	Imperative	Machine code, programming languages	Machine code, Assembly, C

Ultimately, Why/Outcome boils down to “intent” or “will.” What does it look like when that’s generated automatically? Climbing from How up to What and Why, we shift from machine-centric to human-centric communication. One layer higher and even explicit communication fades: your clickstream, biometrics—perhaps every scrap of lifelong data—feeds a super-agent that predicts, proposes, and acts on what you should want next. The input isn't something you write; it's just you.

Layer	Delegates Downward	Describes	Writable Scope	Expression Style	Language Type	Typical Medium (there’s always a spectrum)
Being	Leaves the Why to the super-agent	Unwritten	Unwritten	Unwritten (vast contextual data, incl. activity logs, are the input)	Non-verbal	Your / the business’s very existence
Why	Leaves the What to the agent	Outcome	Why, What, How	Optative	Natural language	Prompts
What	Leaves the How to the runtime	Output	What, How	Declarative	DSLs, programming languages	SQL, Prolog, Terraform
How	Leaves execution to hardware	Procedure	How	Imperative	Machine code, programming languages	Machine code, Assembly, C

We started with the tango between Declarative and Imperative, climbed the recursive staircase to discover Optative Programming, and finally glimpsed a Being layer. Thanks for reading along.

I’d love to hear your thoughts!
If there are any technologies you’re keeping an eye on in the realms of Optative Programming or the Being layer, let me know—imaginative ideas are also welcome!

DEV Community: Yuji Kuroki

Prompts: A Layer Above Declarative Programming?