What is a Bidirectional Expression Resolver?

In a traditional design workflow, expression moves forward. A designer has intent – a color palette, a typographic mood, a spatial rhythm – and translates that intent into design decisions. Those decisions are rendered as components, screens, pages. The intent becomes output.

Design systems and token frameworks support this direction. They encode intent (or at least, values) in a form that can be applied forward – into component libraries, code, production interfaces. This forward-resolution is genuinely useful. It is how brand consistency is maintained across large teams and complex products.

But it is only one direction, and stopping there leaves an important problem unsolved.

The reverse direction – from existing output back to intent – is where the gap lives. Consider what happens when an agent encounters an interface it did not build. Perhaps it is an existing product surface that predates the expression system. Perhaps it is a component contributed by a third party. Perhaps it is an interface a previous agent built without access to the brand's expression identity.

The agent needs to understand what that interface means – not just what values it uses, but what design decisions it embeds, what brand signals it carries, what quality standard it reflects.

Without a resolver that works in this direction, the agent is blind to what already exists. It cannot assess whether a new component will feel consistent with the existing surface. It cannot refine an interface it did not create from scratch. It cannot close the loop between how the brand means to look and how it actually appears in the world.

When a resolver works in both directions, something important becomes possible: coherence as a runtime property rather than a design-time aspiration. The expression system can translate brand intent forward into new outputs. It can also translate existing outputs backward into a brand understanding – reading the current state of a surface and assessing how it relates to the brand's encoded intent.

This is the capability that makes expression infrastructure self-correcting. A brand that evolves – that introduces a new visual language, that shifts its voice, that updates its color philosophy – can apply that evolution not just to new work, but to existing work. The resolver can identify where existing surfaces drift from updated intent and surface that drift to the teams or agents responsible for maintaining those surfaces.

See also: what AI agents need from design for why this reverse-reading capability matters in agent workflows.

Brands do not build surfaces from scratch in one moment. They accumulate surfaces over years – across products, channels, campaigns, and agent-generated interfaces. Each surface was built with whatever expression guidance was available at the time it was built. Over time, that accumulates into a surface estate that reflects the brand's history more than its current intent.

A bidirectional resolver gives brands a way to understand that surface estate – not just to maintain it going forward, but to assess where it stands today relative to where the brand intends to be. This is what turns expression infrastructure from a production tool into a brand governance tool. Not just "here is how to build the next thing correctly," but "here is how everything you have already built relates to your current brand intent."

The direction from output back to intent is what makes the latter possible.

The concept of bidirectional resolution is not specific to brand expression – it appears wherever systems need to maintain coherence between a declared state and an observed state. What makes it novel in the design context is that the "observed state" includes the accumulated aesthetic decisions of designers, engineers, and agents over years of work.

Reading those decisions – understanding what they mean, what quality they reflect, what brand signals they carry – is a form of structural analysis that was not previously possible at scale. The bidirectional expression resolver makes it possible. It closes the loop, which is the prerequisite for any system that needs to stay coherent over time rather than just at the moment of initial creation.

See also: what is expression infrastructure for the broader category this capability serves.