Before Your AI Rollout, Fix This
Governance: Successful AI adoption starts with unified processes and governance, not platform selection or license distribution.
Cleanup: Documentation cleanup exposed conflicting workflows and created the reliable foundation automation needed to connect teams.
Visibility: Integrated systems gave design, commerce, and engineering teams shared visibility into decisions, dependencies, and handoffs.
Guardrails: Narrow AI agents with manual checkpoints reduce dependency risks while preserving human oversight across automated workflows.
Onboarding: Small undocumented tasks, such as printer instructions, revealed valuable knowledge gaps that AI can help distribute.
For a lot of companies, AI adoption starts with a tools conversation: which platform, which license tier, which team gets access first. But according to Jeffrey Cedeno, a Design Program Manager at NBCUniversal who has spent the last year helping roll out AI across design and engineering teams, that's the wrong starting point entirely.
"We launched a governance system for every single operational process that exists so that we can start unifying different teams," Cedeno says. The AI tools came second. Governance came first.
We launched a governance system for every single operational process that exists so that we can start unifying different teams
The “process before platform” ordering is the difference between AI that streamlines an organization and AI that just automates the same mess a little faster.
Create a Free Account to Read More
Unlock this piece and join a community of forward-thinking leaders discovering tools, playbooks, and insights for thriving in the age of AI.
Have an account? Log In
We sat down with Cedeno to get his take on what that's actually looked like in practice on his team: the wins, the guardrails, and the unglamorous cleanup work that made the AI rollout possible in the first place. Here’s what he said.
The Tell: AI Exposes What Was Never Defined
Cedeno didn't set out to rebuild operational documentation. He set out to get design and engineering teams using AI more consistently. But the moment they tried to feed real workflows into AI tools, the cracks showed immediately.
"We found real gaps in our own operation management. For example, one team would have a project plan that they were using and another team would have a completely different document for the same thing," he says. "This wasn't helping with streamlining things or creating continuity."
We found real gaps in our own operation management.
It's the kind of problem that hides in plain sight for years. Everyone assumes their own process is basically fine — until something needs to read across all of them at once, and it turns out there was never one process to begin with.
The Unglamorous Fix Came First
Before any automation rolled out, Cedeno's team did the part nobody wanted to do: they picked one tool for one job, cleaned it up, and killed the redundancy.
"Specifically for us, our SharePoint is really great for live documentation, and that's what it's used for," he explains. "We cleaned up every single folder with very updated information on every single topic for every single team."
Confluence got the same treatment, restructured as a proper wiki with a clear hierarchy. The cleanup wasn't gentle. "I deleted hundreds of pages just so we could restructure all of the data to match our governance system," Cedeno says.
Only after that structure existed did automation start doing real work, which was connecting SharePoint, Jira, and Confluence so that entering a ticket in Airtable could automatically spin up a corresponding Jira ticket, no manual re-entry required. "None of that stuff was possible without the AI integration," he says. But the AI integration wasn't possible without the cleanup first.
The Payoff Was Clarity, Not Just Speed
The most tangible result wasn't faster output; it was cross-team visibility that simply didn't exist before. Cedeno points to the long-running disconnect between design and commerce teams as a clear example.
"We have a commerce team versus a [design] team. These teams work in silos, and now we're all integrated," he says. Previously, a design decision could move forward with no consideration for something like an ad break, because the teams "would just pass things on and we had no idea what happened to it." Now, he says, "every step, this is also somebody to consider, and this is also the team that's involved in this one thing that you don't think about."
The Printer Anecdote
Not every documentation gap is strategic. Some are just glaring once you notice them.
"Even random things — like, how do people use the printer in the company — which was one of the most asked questions in our entire team, was missing from documentation" Cedeno says. "And only two people knew how to tell you how to do that."
Even random things — like, how do people use the printer in the company — which was one of the most asked questions in our entire team, was missing from documentation.
It's a small example, but it's the clearest illustration of the actual problem: undocumented knowledge, sitting in two people's heads, that was previously seen as too trivial to put in documentation. But it’s pieces of knowledge like that that can make AI so much more useful, especially for newer people in an organization leveraging it for onboarding.
The Guardrail: Narrow Agents, Not One Mega-Brain
For all the enthusiasm, Cedeno is notably cautious about over-consolidating. His team isn't trying to build one system that does everything.
"We're trying to create very specific agents and use cases and workflows, and then limit it there," he says. A single system that handles everything sounds efficient, but he sees the tradeoff clearly: "It does create a situation where you're now dependent on this one thing, and the less human interaction you have, the more there's a capacity for things to go to hell."
That's why every automated workflow at NBCUniversal still has what he calls a manual checkpoint. "There's a human interaction level, like a bypass key, across every single touch point," he says, "so that if something's not right, someone's going to catch it."
The Bottom Line
Cedeno's experience is a fairly direct rebuttal to the way most companies talk about AI adoption. The instinct is to lead with the tool: get everyone licenses, run some workshops, wait for productivity to climb. What actually moved the needle at Peacock was slower and less exciting: agreeing on one process, cleaning up the documentation behind it, and only then letting AI automate what was already well-defined.
"It's never going to be perfect," Cedeno says. "I can definitely tell you that it's helped us make it as perfect as it can be for now."
Want more insights like these? Sign up for a free DPM account to hear from more experts like these.