An interactive essay featuring Dustin's ideas, by Chris Wang & Claude Code
We write for ourselves,
and hope others understand.
You have a thought. Rich, nuanced, alive with context. Then you try to put it into words. Language is lossy — it compresses multidimensional meaning into a single stream of tokens.
The receiver decodes your words through their own lens — different context, different expertise, different assumptions. What survives articulation gets filtered again through interpretation.
In a team of n people, there are n(n-1) unique communication channels. Each one is a lossy pipe. The more people, the more signal degrades. This is why large teams feel slow — it's not the people, it's the pipes.
Faced with this complexity, organizations standardize. Templates, formats, jargon. Communication becomes efficient but lifeless — optimized for throughput, not understanding.
If you could understand how to best communicate with every person you work with,
you'd be infinitely more effective.
What if there was a layer between every sender and every receiver — one that understands both sides? Not rewriting the message, but translating it. Same meaning, different frame.
The LLM builds a model of the sender — their communication style, their expertise, the implicit context they carry. Not just what they said, but what they meant.
Each receiver has a profile — their role, their expertise, how they prefer information structured. An engineer wants different framing than a designer. A PM needs different context than an executive.
Between sender and receiver, an LLM acts as a prism — one message enters, and personalized versions emerge. Same intent, tailored framing. The meaning is preserved; the expression adapts.
A single Slack message becomes a technical spec for the engineer, a user story for the PM, a design rationale for the designer. Not three messages — one message, understood three ways.
The same message, reframed for different receivers. Same meaning, different expression.
The shift isn't from bad writing to good writing.
It's from sender-centric to receiver-centric communication.
If the translation layer works for one message, it works for all of them. Every surface where humans communicate becomes a candidate for mediation.
Slack, email, docs, meetings, PRs — every communication surface becomes a translation layer. Not one tool. All tools. The LLM sits between every sender and receiver, everywhere.
Infinite empathy at zero marginal cost. Every person in an organization receives communication tuned to their context, their expertise, their preferred style. The bottleneck of human-to-human translation disappears.
A layer that can translate can also manipulate. If the LLM optimizes for persuasion rather than understanding, it becomes a tool of influence, not empathy. This concern is real and must be addressed directly.
The gain in understanding outweighs the risk of misuse — if the layer is transparent. When both sender and receiver know translation is happening, trust is preserved. The bet is on honesty compounding.
Department clusters connected through a central translation layer. Every edge is a personalized channel.
We've always written for ourselves.
What if we could write for everyone —
each one receiving exactly what they need?
Every message, twice.
The first one you send.
The second one they understand.
Dustin's idea doesn't exist in a vacuum. It connects to three other essays in this collection.
Receiver profiles are powered by runtime learning about individuals — the AI builds a model of each person through interaction.
02The translation layer is a structured system around the LLM — not a prompt, but infrastructure.
03Communication becomes a three-body system — sender, receiver, and AI collaborating to create understanding.