How a media newsroom scales judgment, not volume

Code & Theory, the agency behind NBC's election night big board and the Minnesota Star Tribune's digital reinvention, argues that media success now depends on editorial judgment, not content volume, as generative AI has made plausible articles cheap and abundant. In a webinar, Michael Fanuzzi, head of immersive technology, and Rebecca Smith, senior director of product, detailed their 'intelligence layer' system, piloted for the upcoming true crime publication RealClear Crime, which places human editors in the middle, surfaces uncertainty in fact-checking, and adds intentional friction to preserve integrity. 'AI has really put the final nail in that coffin,' Fanuzzi said, referring to the old volume-based model.

How a media newsroom scales judgment, not volume Inside the structure, workflows, and standards that let editors move faster without lowering the bar. Hollie Aghajani Staff Product Marketing Manager Published For two decades, the media business ran on one equation: more content, more page views, more revenue. Sanity partner, Michael Fanuzzi, head of immersive technology at Code & Theory, traces the habit to the 24-hour news cycle, when there wasn't always enough happening to fill the airtime but it had to be filled anyway. When news moved online and the economy became page views, that logic multiplied. The tools built in that era were tuned for those levers, because it was what most publishers had. Fanuzzi is blunt about what finally broke the equation. "AI has really put the final nail in that coffin." The scarcity the old model depended on is gone. "We don't want for more content anymore. If anything, we're drowning in it." Run the old volume playbook with new generative tools and you don't get an edge. You get slop, produced faster than anyone can read it. This was the subject of the second episode in our Structured Series of webinars /events/structured-series-ai-media , a conversation with Michael Fanuzzi and Rebecca Smith, senior director of product at Code & Theory, the agency behind NBC's election night big board and the Minnesota Star Tribune's digital reinvention. Their thesis: the future of media won't be won by whoever publishes the most. It will be won by whoever publishes with the best judgment. Why volume stopped being a strategy Here's the mechanism most coverage of AI in newsrooms misses. Generative AI moved two lines at once. It raised the ceiling on what a small team can produce, and because it did that for everyone, it raised the floor on what audiences expect. When anyone can generate a plausible article in seconds, a plausible article is worth nothing. "Volume alone is not a differentiator anymore," Fanuzzi said. "Quality comes from judgment, which is the unique thing that our editors and journalists bring to the table." Where human judgment belongs Code & Theory's answer is a system they call the intelligence layer, first built for RealClear Crime, a new true crime publication launching this fall. The pilot was chosen on purpose. A new publication in 2026 with a small team, big ambitions and every pressure Fanuzzi named, minus the editorial army only a few legacy titles can still afford. If judgment can scale anywhere, it has to scale there. Three design decisions of the intelligence layer where a human belongs in the work. The human sits in the middle. The system organizes the Fact-checking surfaces doubt instead of burying it. The fact-checker breaks a draft into individual factual questions and corroborates each against secondary sources. When it hits a claim it can't confirm, it flags it rather than waving it through. The system can't interview a source or knock on a door, and it doesn't pretend to. What it can do is tell an editor exactly which claims still need a human to go report them. "It's surfacing its uncertainty rather than hiding it," said Fanuzzi. That's the opposite of how most of us have experienced AI writing tools. The friction is intentional. Most product work is about removing friction. This system adds it: an editor has to mark each stage reviewed before the next unlocks. "Journalists want to live in the truth, and the idea of handing off some of that integrity to AI is a little bit scary," Smith said. "They wanted friction. This was a user-informed choice." The same instinct answers the hardest cultural problem in adopting these tools, the gap between leadership mandating AI and writers who distrust it: position the tool as enablement, not replacement, and build it on sources people can verify. Watch the demo of the intelligence layer below. Trust is earned in the workflow, then spent at publish Trust in an AI-assisted newsroom starts before any AI runs, with the organization agreeing on what it considers true. "It's incredibly important to have a shared definition of what the truth is, alignment on what sources can be trusted," Smith said. Generate a draft off the open internet and an editor gains nothing, because they'll re-verify everything anyway. The gates do more than catch errors. They're what lets an editor believe in the piece in the first place. Every stage that has to be reviewed, every claim that gets flagged, is a chance for a person to satisfy themselves that the story holds up, so that by the time it's ready to publish they actually trust it. Journalists are the hardest possible audience to sell that exchange to, and Fanuzzi treats it as a feature. "Their entire job is to not be credulous. They are skeptics by nature." A tool that hides its uncertainty loses that room immediately. A tool that surfaces it, that shows its sources and flags what it couldn't confirm, earns the benefit of the doubt because it behaves the way a good editor already does. The exchange runs all the way to the reader. On the published page, the system marks each piece for what it is: AI-assisted, edited by a named human who stands behind it. Transparency was a founding design principle, Smith said, because a reader can only extend trust to a newsroom that shows its work. The volume era asked audiences to trust the byline and never see the process. This inverts it, where the process is the proof. Faster is linear. Compounding is the moat. If you take one idea from the session, take Fanuzzi's distinction between a system that's fast and one that compounds. "Faster is linear. Everyone has these tools now. Anyone can spin up a plausible-looking content site in a day, and that's no longer impressive." A compounding system doesn't stop at publish. "A pipeline ends at publish," Smith said. "That's where our loop starts to do the work." Engagement flows back as insight, surfacing what a dashboard can't: an uncovered trending gap, older content still drawing readers, a dormant story ready to resurge. "What your competitors can't replicate is your bespoke insights," Smith said. Anyone can buy the same SEO tool. No one else has your audience correlated against your archive. The goal shifts from clicks to reader relationships. What this asks of a newsroom now Give your original reporting the same scrutiny as everything else. Asked whether fact-checking applies to proprietary reporting or only to aggregated content, Smith's answer was one line: it runs on every story. That challenges a common bias. Newsrooms tend to trust their own reporting reflexively and reserve verification for outside material. In a system where AI can corroborate any claim against secondary sources, your own bylines don't get a pass. Treat the AI draft as a floor, not a draft. Asked what kind of content this really helps with, Smith offered the most useful reframe for any editor nervous about generative tools: the system's job is to kill the blank page. It produces the generic, get-you-started version so a content owner never stares at nothing, and the editorial value and the judgment about what matters, is entirely in what you do next. If your writers are shipping the floor, you've bought the wrong tool. If they're standing on it, you've freed them for the work only they can do. Stop trying to store the truth. Track it. There's no single canonical database of "true facts," Fanuzzi said, just recent, authoritative sources superseding older ones as new references appear. That cuts against the instinct to verify a fact once and file it away. In a live news environment, facts get superseded: a newer record, a better source, a correction. So build for provenance. Track where every claim came from and when, so re-verification is cheap when sources move. Truth isn't stored once. It's a chain you keep current, or as the team put it, made traceable. So here's the question to take back to your own publishing. The scarce thing was never content, and it certainly isn't now. It's the judgment about what deserves to exist, whether it's true, and whether it's worth your readers' time. You can spend your next tooling budget making more of what everyone already has, or you can spend it defending the one thing a competitor can't replicate. Which one you choose is the strategy. This piece is part of Sanity's Structured Series, where we talk with leaders reshaping how their organizations create, manage, and run AI