Problem
An editorial gap traditional newsrooms couldn't fill.
Younger audiences have migrated to vertical formats, soft-news content and a continuous publishing rhythm. The structure of a traditional newsroom — human anchor, physical set, fixed schedule — doesn't scale to the frequency or cost required to compete on that ground. The question wasn't whether it made sense to experiment with AI, but how to do it without breaking the editorial line of a national media company.
Context
Grupo Fórmula, 2023 — before the mainstream generative AI boom.
NAT launched in March 2023 — before "generative AI" was boardroom vocabulary. At the time there were no playbooks, no frameworks, no precedent cases in the region. Every decision — platform, governance, editorial line — was made under genuine uncertainty, not the comfort of copying something someone else had already solved.
Decision
Not an experiment. Production, with governance from day zero.
The executive decision was to treat NAT as a product in production, not a demo. That means: continuous presence across platforms, an editorial calendar, an identified human owner, a review procedure before publishing, and internal policy aligned with the group's editorial line. No "let's put it out there and see what happens."
Architecture
Modular pipeline: voice synthesis, visual generation, orchestration and publishing.
Four independent layers that can each be updated without taking down the system: (1) script generation with human editorial review; (2) voice synthesis trained specifically for NAT; (3) visual generation and animation; (4) publishing across channels on a scheduled cadence. Each layer has its own approval threshold and its own audit point.
Execution
From concept to air in under one quarter.
Shipping to production wasn't the interesting milestone — the interesting part was keeping the operation stable over the following months, iterating on real audience feedback, and sustaining editorial accountability without letting the novelty become an excuse to lower the standard.
Result
Continuous presence, international coverage, and an institutional precedent.
NAT is still in production three years later. Her public account (@nat_tvoai) remains regularly active. The case led to an editorial interview with the Reuters Institute (University of Oxford) — and outgrew its original journalistic use into brand-content formats across the region.
Lesson
The barrier wasn't technological. It was institutional.
The tools to build NAT already existed — imperfect, but sufficient. The real barrier was institutional: convincing a board that the reputational risk was worth taking, aligning Legal, the newsroom and Production around a product the industry didn't yet have a name for, and defining an accountability framework when the "anchor" isn't a person. That's still the real blocker on almost every project that reaches a CAIO's desk.
Governance
An identified human owner, always.
Every piece NAT publishes has an accountable human editor. The approval process isn't skipped for urgency or volume. When a gray area comes up — a sensitive topic, a story that could read two ways — the default is to escalate, not to publish fast. That discipline is what makes the operation defensible under outside scrutiny.
External evidence
Third parties who documented the case.
-
Reuters Institute · University of Oxford
"Meet NAT, an AI-generated presenter offering soft news to Mexican audiences"
View publication → -
Reuters Institute — research program
AI and the Future of News / Journalism
View program → -
Latin American Journalism Review · Knight Center
Media outlets, universities and governments experiment with AI in news anchors
View article (Spanish) → -
LinkedIn News
"She's NAT — the first AI-created TV anchor in Latin America"
View story (Spanish) → -
Universidad Diego Portales · Vergara 240
Interview on AI and journalism
View interview (Spanish) →
Related