Fall '26 introduced a new theme architecture that moves page structure out of JSON templates and back into Liquid. On the surface that reads like a reversal of Online Store 2.0.
It is a reversal — of the optimisation target, not of the progress. OS 2.0 optimised for a visual editor, so structure had to be machine-editable JSON. The new architecture optimises for a reader that consumes a page from top to bottom, and that reader is increasingly not human.
Key takeaways
- Page structure returns to
templates/*.liquid. A new{% block %}tag mixes with plain HTML. - Typed contracts on block inputs; a
{% partial %}tag lets JavaScript refresh fragments. - Tailwind is built in and only the CSS a page actually uses is shipped — a direct Core Web Vitals win.
- Themes carry markdown companion files that explain to Sidekick how the theme is built.
- Developer preview since July 2026, running alongside existing themes. No forced migration, no announced GA date.
Old and new, side by side
Dimension | Online Store 2.0 | Fall '26 architecture |
|---|---|---|
Page structure |
|
|
Components | Section plus nested block schema | A |
Type safety | Loose schema conventions | Typed contracts on block inputs |
Partial updates | Section Rendering API | A |
Styling | Bring your own, shipped site-wide | Tailwind built in, only the CSS a page uses is sent |
Machine readability | Rebuild the page from JSON plus a section tree | One file, read top to bottom, plus a markdown companion |
Why JSON was the problem
To understand a product page under OS 2.0, you parse a JSON template listing section references in order, resolve each to a Liquid file, walk a nested block schema, and reconstruct the result mentally. For a visual editor that is ideal — every node is addressable.
For a language model it is expensive and lossy. Context is spent on reassembly rather than comprehension, and the reassembled picture is frequently wrong. Anyone who has asked an AI assistant to modify a theme has seen the failure: it edits the right-looking thing in the wrong place.
A linear Liquid template removes the reassembly step. Humans and agents read the same file in the same order and arrive at the same understanding.
Three consequences beyond AI
Performance. Shipping only the CSS a page uses attacks one of the most durable problems in Shopify themes: a global stylesheet carrying rules for sections the page never renders. This lands directly in LCP and CLS. For stores where theme bloat has resisted every other fix, it is the first structural remedy in years.
Fragment updates without the API round trip. {% partial %} lets JavaScript refresh a named fragment — cart counts, price blocks, stock badges — without the Section Rendering API choreography. Fewer moving parts, fewer cache bugs.
Typed contracts. Block inputs get types. Mis-typed settings fail at the boundary rather than rendering something subtly wrong three screens down.
The markdown companion is the real tell
Themes under the new architecture ship markdown files describing how the theme is built, written for Sidekick to read.
That is a genuinely new category of artefact: documentation that is not for a developer joining the team, but for an agent about to edit the code. It is the same instinct as /agents.md at the store level, applied to the theme — and it is why Canvas, the single-view store editor, can let a merchant redesign a theme by describing what they want.
It also implies a new obligation. If you build themes for clients, that file is now part of the deliverable, and its quality determines how well the merchant's AI assistant can work with what you built.
What to do about it now
If you run a store. Nothing urgent. The architecture is in developer preview, runs alongside existing themes, and has no GA date. Ask your theme partner whether they are tracking it, and keep the question for your next major refresh.
If you build themes. Start now. Preview has been open since July 2026, and the first wave of migration demand will arrive concentrated at GA. The CSS subsetting alone is a sellable performance story before any AI argument.
If you are weighing headless. This narrows the gap. A notable share of headless migrations were motivated by escaping theme-layer performance and structural limits. Tailwind with per-page CSS, typed blocks and partial updates address several of those directly. Combine that with Hydrogen becoming a library rather than a framework and the binary choice between Liquid and headless has turned into a spectrum.
FAQ
Will my current theme stop working? No. The new architecture runs alongside existing themes with no forced migration and no announced deprecation.
Does it improve AI visibility by itself? Indirectly. It makes a page easier for an agent to read and understand, which helps the citation layer and makes AI-assisted editing reliable. It does not change what the commerce tool path returns — that comes from catalog data, not markup.
Is Tailwind mandatory? It is built in. Treat it as the default rather than a requirement, and evaluate how it fits your existing design system before committing a rebuild.