Problem
Several teams shipped into the same retail surfaces, so every deploy depended on everyone else, and performance regressions on high-traffic pages were expensive at Mercado Libre's scale.
The shape of a deploy
The problem was never the code. It was that every team’s change waited for every other team’s change, and when something broke on a page half of Latin America was looking at, finding the owner took longer than the fix.
Four teams, one shared release. Anyone's change waits for everyone's, and anyone's bug is everyone's page. Four teams, four pipelines. A change reaches users when its owner says so, and a bug pages exactly one phone.
The work was in the seams: one owner per slice of the page, a thin shell that owned routing, design system and telemetry, a contract small enough to fit in your head, and failure isolation so a slice that did not load rendered its fallback while the rest of the page kept working.
How the standards travelled
Independent deploys gave teams autonomy. Performance budgets in the pipeline kept that autonomy from turning into slow pages. Neither would have held without the same few review comments, repeated across teams until they were habits.
apps/gallery/src/render.ts changes requested
14
import { loadRecommendations } from "./recommendations"; 15 +
import { trackingScript } from "@vendor/analytics"; 16
17 +
document.head.appendChild(trackingScript()); 18
export async function renderGallery(productId: ProductId) { 19 +
const recommendations = await loadRecommendations(productId); Recommendations belong to Discovery. Gallery should not wait on their slice. Load it in their rectangle, not ours.
20
const gallery = await loadGallery(productId);
Who owns this script, and why is it on the critical path? If analytics needs it, it loads after first paint.