A serious lab does not become serious by sounding abstract.
It becomes serious by making claims that point somewhere: a product, a paper, a repository, a method, a working surface, a system that can be inspected.
This is the design rule behind CDLI's public site.
The visual language should be quiet enough to let the artifacts carry weight. The writing should be direct enough that a technical reader can tell what is real. The structure should be stable enough that future pages feel authored by the same institution.
Operating rule
If a public claim cannot point to an artifact, it should be rewritten or removed.
That rule is useful for branding, but it is also useful for engineering. It prevents the site from drifting into broad language while the actual work remains specific.
The problem with polish
Polish is not the enemy. Empty polish is.
A young company needs a face. It needs typography, rhythm, logo usage, language, and enough restraint that the work feels intentional. But if polish arrives before proof, it creates a dangerous illusion: the surface gets better while the institution stays hard to understand.
That is especially risky for an AI lab.
The market is full of fluent language. Anyone can write that they build intelligent systems, transform operations, or unlock value. Those sentences are cheap because they do not make the reader smarter. They do not reveal how the lab works. They do not create trust with an engineer, founder, operator, or future collaborator.
Proof does.
Proof can be small. A raw Markdown route is proof. A source repository is proof. A method with a trigger and output format is proof. A working product surface is proof. A diagram that maps a system before judging it is proof.
The public site should make those objects easy to find.
What proof looks like on a website
For CDLI, proof has a few forms.
- 01A product or collaboration that names what exists.
- 02A public method that shows how work is done.
- 03A paper or note that explains the operating thesis.
- 04A machine-readable surface that agents can index.
- 05A contact path that does not hide behind vague positioning.
Those objects do not need to be loud. They need to be inspectable.
This is why the site uses a restrained visual system. Graphite surfaces, mono indexes, red action signals, and the CDLI seal are enough. The brand should feel like an institutional command surface, not a template for an AI agency.
When the page gets too decorative, the proof becomes harder to see. When the page is too plain, the institution loses presence. The right place is in the middle: serious, sharp, and artifact-led.
A useful test for new sections
Before adding a new section, ask:
- -What artifact does this point to?
- -Would a technical reader learn something real?
- -Would an agent be able to index or cite it?
- -Does the language say what exists, or only what we want to signal?
- -Does the design reuse the system, or invent a new mood?
If the answer is weak, the section should wait.
That is not minimalism for its own sake. It is governance. Every public surface teaches future contributors what kind of company this is.
Why this matters early
Founding a company is partly an act of choosing what not to become.
CDLI should not become a collection of disconnected pages, random visual effects, and inflated claims. It should become a lab whose public face has memory: same mark, same structure, same evidence bias, same refusal to sound bigger than the work.
Proof before polish does not mean ugly before beautiful.
It means the beauty has a job.