Syntax is overhead
Braces, boilerplate, naming conventions, and layers of abstraction were invented to make machines legible to people.
A language for the next runtime
The programming language built for machines that write code.
Humans describe intent. AI compiles reality. LinkedSpec is a new kind of programming language designed for the moment software is authored by intelligence.
Optimized for Claude models · Open to every model
INTENT FIRST ✦ MACHINE NATIVE ✦ TOKEN EFFICIENT ✦ CLAUDE COMPATIBLE ✦ INTENT FIRST ✦ MACHINE NATIVE ✦
01 / The premise
Code was built for human readers. That era is ending.
Every major language carries the same assumption: a person will write it, read it, and maintain it. But the author is changing. AI can reason over structure directly, without the ceremony we created for ourselves.
Intent is the source
Give an AI the outcome, the boundaries, and the constraints. Let the compiler choose the representation that fits the machine.
Efficiency becomes native
Fewer tokens in the instruction. Less data across the wire. Smaller artifacts on disk. The language can optimize all three.
02 / See the difference
Describe the outcome.
Skip the ceremony.
LinkedSpec is deliberately compact. It gives intelligent systems a precise surface for intent, constraints, and events — then gets out of the way.
−64%target token volume
−41%target transfer size
1:1intent to execution trace
01
intent: ship a realtime workspace02
03
workspace {04
auth: passwordless05
sync: realtime06
presence: live07
storage: durable08
09
when member joins room:10
grant access(room)11
broadcast presence(member)12
}03 / The loop
One intent.
Four precise moves.
Intent
State what should exist, happen, or remain true. Natural language is welcome at the edge.
Constraint graph
Define the boundaries: security, latency, data, permissions, and the conditions that cannot be violated.
Execution plan
LinkedSpec resolves dependencies and selects the smallest viable plan for the target runtime.
Machine output
Emit the most efficient representation for the environment: edge, cloud, device, or whatever comes next.
04 / Built with Claude in mind
Give Claude a language that speaks directly to the machine.
LinkedSpec is model-compatible by design and deeply efficient with Claude. Sonnet can shape intent, Opus can reason through complex constraints, and Haiku can transform small instructions at speed.
See the model guide ↗Claude Opuscomplex systems · architecture
01Claude Sonnetdaily builds · balanced output
02Claude Haikufast transforms · edge tasks
03One compact language.
A whole family of capable minds.
A whole family of capable minds.
05 / The manifesto
“Humans describe intent.
Machines compile reality.”
We are moving from a world where people write every instruction to one where people define the boundaries of what should happen. LinkedSpec is the syntax for that transition: explicit enough to trust, compact enough to move, and native to the intelligence doing the work.
Continue reading ↓06 / Field notes
Ideas for the machine-made era.
01
The End of Human-First Syntax
Programming languages were shaped around the limits of human readers. AI changes the reader, the author, and the economics of every symbol.
Read article ↗02Why AI Needs a New Programming Language
A language for machine authors does not need to explain every step to a person. It needs to preserve intent while removing avoidable work.
Read article ↗03Tokens Are the New Runtime Cost
When models write software, tokens become build time, network traffic, and operational cost. The representation itself becomes an engineering decision.
Read article ↗07 / The first instruction
Tell the machine
what matters.
LinkedSpec is early, intentional, and open to builders who believe software should be described at the level of purpose.
Try the syntax ↗Questions, anticipated
Make the shift.
LinkedSpec is an AI-first programming language designed around intent, constraints, and execution. People describe outcomes; an AI compiler resolves the most efficient machine representation.