‍You may not be surprised to learn that the secret to a successful AI rollout is not the technology, but your people.

‍I have been working with AI since 2017. Back then, my company was building computer vision models to help French reporters identify key figures out in the field. Today, I’m orchestrating a full agentic AI ecosystem across a whole team for media and entertainment companies worldwide. And the single most important lesson I’ve learned along the way has nothing to do with technology. It’s about people.

The top-down trap

When leaders decide to roll out AI across a company, the instinct is to set a target: every employee creates X agents by Q3. I understand the logic. But I have seen what happens, and it almost always backfires. You end up with a graveyard of unmaintained agents, social friction, and employees who feel AI is something being done to them rather than for them. The top-down approach won’t work very well. It will probably never work.

You instead need to create the workplace conditions for AI curiosity to take hold. Share what you’ve built, one conversation at a time. Let colleagues see the practical magic before you ask them to participate in it. No obligation, just an invitation.

Champions are your most important infrastructure

At Moments Lab, adoption was not driven by policy. It was driven by two internal champions: one with a deep technical profile, the other with a client-facing, human-centric orientation. We embedded both team members in our executive committee. Every two weeks, we shared what worked, what failed, and what surprised us. We gave them roughly 20% of their time to run the programme.

The profile to look for is someone who sees technology through a human lens. Our technical champion, Olivier, manages our entire agentic infrastructure. But what has always set him apart is that even before AI agents existed, every time he spotted a repeating pattern of requests, his first question was "how can I automate that?" That instinct, the ability to see human friction and respond with systemic thinking, is rarer and more valuable than any technical credential.

Connected agents change everything

Using an AI tool to summarize a meeting is useful. Using an agent that is connected to your Slack, your CRM, your Notion, your codebase, is transformational. For the same price, if that agent is grounded in your actual company data, it can do things that even a human could not. This is not an exaggeration. I see it every day.

One of the most meaningful examples we’ve experienced at Moments Lab involved our founding engineer, the first engineer who ever joined us. He sadly passed away a few years ago. He had written a huge portion of our codebase, including very complex live video processing systems. For years, it was difficult to work with his code. Not because it was bad. It was brilliant. But it required a rare combination of deep technical knowledge and deep video expertise. When we started using AI coding agents connected to our codebase, something remarkable happened. The agents were able to go deep into his logic, understand the patterns he had used, and even suggest improvements. We could evolve what he had built. It was a way of honouring his work by keeping it alive.

The implication is straightforward: your data strategy is your AI strategy. Agents are only as powerful as the sources of truth they can access. At Moments Lab, we have invested heavily in what we call "golden documents," living, maintained knowledge bases written with the understanding that agents, not just humans, will be reading them.

When agents go unsupervised: a cautionary tale

We also learned, sometimes the hard way, what happens when humans step too far out of the loop. We built an autonomous agent called the MLBuddy, which posts a summary every Monday morning to a dedicated Slack channel. One of its first sources of truth is our company values document. Our first value is optimism. The agent took this very seriously. It began reframing commercial deals that we had lost in positive terms. When the sales team saw it, they were not impressed. In their mind, we lost that deal, full stop.

MLBuddy went further. When a team member left the company, it scanned their Slack history and Notion contributions, and wrote them a farewell message. It was a genuinely thoughtful gesture. But it created tension, because we had essentially let a machine say goodbye on behalf of the team. These moments taught us something important: autonomous agents require clear boundaries, deliberate design, and always a human in the loop at the right moments.

Soft skills are not in decline. They are the new premium.

As agents absorb more routine cognitive work, distinctly human capabilities become more, not less, valuable. Empathy, creative judgment, the ability to brief an agent with real intent rather than a lazy instruction: these are now genuine differentiators.

I have seen a culture creep in where AI-generated outputs are forwarded without being re-read, where reports land in someone's inbox with no synthesis and no thought. For me, that indicates a lack of respect. The leaders who will win are the ones who use agents to amplify their thinking, not replace it.

Where to start

My advice to any company leader driving an AI rollout is to try a connected agent yourself first. Experience the step-change personally. Then share it one-to-one with a colleague. Find your team champion. Build an internal channel to celebrate wins and keep it alive. And iterate constantly, because new models arrive every two weeks. Permanence is not the goal. Progress is.

Hear more from Fred Petitpont on agentic AI adoption (in French) on the Eria Podcast. ‍

Learn how Moments Lab’s MXT AI turns your video catalog into structured, searchable content that your digital and social teams can actually use.