The token pipeline almost nobody actually built

You changed a colour in Figma this week. Then you opened the repo and changed it again. Nobody wrote that step down anywhere, and nobody will, because it only takes a minute and you already know both places.

The format stopped being the excuse

For years the honest answer to "why isn't this automated" was that every tool spoke its own dialect. That expired on October 28, 2025, when the Design Tokens Community Group shipped the first stable Design Tokens Specification, version 2025.10, backed by more than 20 editors and authors from Adobe, Google, Microsoft, Figma, Salesforce and Shopify.

A token has a value and a type, and tokens can point at other tokens. That is most of it, and it is deliberately boring.

A shared format was the hard part to agree on and the easy part to adopt. Once it existed, moving a token from a design tool to a codebase stopped being an integration project and became a file.

Both ends exist, the middle does not

zeroheight surveyed 147 design system practitioners for its fifth annual report this year. Both ends of the pipe are in place almost everywhere: 86% include design tokens in their system, 90% hold the definitions in a design tool, 82% hold them in code.

The middle is not there. 60% have no token automation at all. Only 40% have any pipeline established, which means most teams carrying tokens at both ends are moving them across by hand.

That is a strange shape for a problem to hold. The endpoints are near-universal, the standard is stable and vendor-neutral, and the connection between them is still a person remembering.

What a hand-synced token actually costs

Not much, once. That is the whole trap.

The cost is not the minute you spend. It is that a manual step has no failure signal. A broken build tells you it broke. A colour you updated in one place and not the other tells you nothing, and keeps telling you nothing until a designer opens a screen six weeks later and finds two greys that were supposed to be one.

By then the question is no longer which value is right. It is which of the two has been shipped to customers, and for how long. A manual sync does not fail loudly, it fails quietly and then gets inherited.

There is a second cost that shows up in the same survey. 66% of teams have documentation for their tokens, which is higher than the 40% who have a pipeline. More teams have written down what the tokens mean than have automated where they go. Documentation is the thing you produce when the process cannot be trusted to explain itself.

Decide which end is the source

The reason this does not get automated is rarely the tooling. It is that nobody has made one decision, and the decision feels bigger than it is.

The decision is direction. Either the design tool is the source of truth and code receives, or code is the source and the design tool receives. Not both. A system where both ends can write is not a pipeline, it is two sources of truth and a merge conflict waiting for a person.

Most teams avoid choosing because each direction has an obvious loser. Designers lose the ability to change a value and see it immediately if code owns it. Engineers inherit whatever a design tool exports if design owns it. Both objections are real, and neither is as expensive as not deciding.

The twenty minutes that make this fixable

Before any tooling, settle the direction out loud and put it at the top of the design system README. One sentence, about twenty minutes of argument, no install.

Colour and spacing tokens are defined in Figma. Code receives them and never writes back. your name syncs by hand on Fridays.

Two things change immediately. A disagreement now has a winner, so nobody relitigates it in a pull request. And the receiving end is on record as the place that gets overwritten, which is what people quietly resent later when it was never said out loud.

The clause most teams leave out is the last one, and it is worth more than the rest. "By hand, on Fridays, by me" is not an admission of failure. It is a cost with an owner, written where someone other than you can read it, and owners get budget in a way that invisible steps never do.

The format has been settled since late 2025. What is left is a decision about direction that no specification can make for you, and that most teams are still quietly paying for by hand.