The following is from a recent Radical Ventures AI Founders Masterclass session with Cohere Chief AI Officer, Joelle Pineau

Artificial intelligence is advancing at extraordinary speed, yet one of its most persistent challenges remains unsolved: the leap from research to real-world impact. Models continue to grow in capability, but knowing when they are ready, and how to prepare them for the complexity of production environments, is still one of the most challenging transitions in applied AI.

Few people understand this transition more deeply than Joelle Pineau. Over the course of her career, from professor at McGill and co-director at Mila, to leading FAIR at Meta, and now serving as Chief AI Officer at Cohere, Joelle has worked at every layer of the AI ecosystem. Her leadership has shaped research labs, product teams, and the systems that translate frontier science into practical value.

In her Masterclass conversation with Radical Partner Sanjana Basu, Pineau offered a rare look at the internal realities of applied AI: the judgment calls, the cultural foundations, and the organizational mechanics required to move breakthrough research across the threshold into meaningful deployment.

Bridging the Gap in Applied AI

For Pineau, the hardest part of applied AI isn’t the modeling, it’s the judgment. Benchmarks can only go so far. Data can mislead. And even rigorous evaluations can’t fully predict how a model behaves outside the lab.

What bridges that gap is something more human: the conviction to take responsible risks, informed by deep context and guided by lived experience.

During her time at FAIR, Pineau developed what she calls the “graduation model” — a mechanism for carrying ideas from research into product not just as code, but as knowledge. Instead of tossing a model over a wall, teams brought researchers with the work. Tacit understanding, practical nuance, and accumulated intuition traveled alongside the technology.

“Evaluation frameworks help,” she notes, “but at some point, someone has to decide the research is ready, and be prepared to run with it if it works.”

This way of thinking reframes applied AI not as a linear pipeline, but as a dynamic continuum. One that demands flexibility, humility, and leadership that can operate in uncertainty.

Culture as Infrastructure

Pineau’s time at FAIR also revealed a different kind of scaling challenge, one that had little to do with compute. As the lab grew from 100 to more than 600 researchers across ten global locations, techniques and strategies evolved rapidly. Yet one thing remained stable: the culture.

Rooted in a shared set of values, FAIR’s culture became the invisible infrastructure that allowed the organization to navigate ambiguity, tension, and high-stakes decisions.

“Culture scales faster than technology,” Pineau reflects. Values shaped collaboration, informed trade-offs, and became a stabilizing force in a landscape that shifted constantly. They anchored the team through years of scientific disruption and organizational transformation.

In Pineau’s view, culture isn’t a nice-to-have. It’s a strategic instrument, one that determines whether a research organization can adapt and sustain excellence over time.

The Plus-One Principle: A Framework for High-Trust Leadership

One of Pineau’s defining leadership philosophies is what she calls the Plus-One Principle. When leaders enter a room, they aren’t there to defend their function or team, they’re responsible for the success of the entire organization.

This simple shift changes everything. It removes zero-sum thinking, reduces territorial friction, and accelerates alignment. It creates space for productive tension and allows teams to wrestle with hard decisions without losing sight of shared purpose.

At FAIR, Pineau applied this principle in rooms ranging from lab leadership meetings to Meta’s executive committee. At Cohere, she brings it into the next generation of enterprise AI, where cross-functional trust is essential for building systems that must operate reliably at scale.

Rigor as a Strategic Advantage

Pineau is also a long-standing champion of scientific rigor. Early in her academic career, she and her students attempted to reproduce widely accepted reinforcement learning results and failed. That experience sparked a years-long movement to establish better methodological standards across the field.

Her stance is direct: rigor and speed are not opposites. Rigor is what makes speed meaningful. It ensures that breakthroughs are durable, that knowledge compounds rather than collapses, and that teams build on foundations they can trust.

In an industry that often rewards shipping fast, Pineau argues that reproducibility is itself a competitive advantage — especially in high-stakes domains where errors scale quickly.

Toward the Next Frontier of Applied AI

When looking ahead, Pineau sees momentum building in several areas that sit at the intersection of deep science and real-world demand. Scientific and materials discovery is accelerating at a pace that was unthinkable a decade ago. Robotics is regaining energy as researchers push toward systems that can operate robustly amid real-world variability.

Pineau is particularly focused on the rise of agentic AI, systems capable not just of understanding language, but of taking actions, orchestrating tools, and driving enterprise productivity. These systems introduce new responsibility challenges, but also enormous potential for impact.

And while Silicon Valley remains a center of gravity, Pineau is equally energized by what’s happening in Canada’s AI ecosystem. After years spent moving between Montreal and California, she sees this as a defining moment for the region, a time to build boldly, learn quickly, and embrace the value of “intelligent failures” that strengthen the next generation of founders.

Building the Next Era of AI

Across Pineau’s career, one theme emerges again and again: breakthroughs do not yield impact without leadership. The translation from research to the real world depends on clarity, trust, rigor, and the courage to make hard decisions with imperfect information.

It requires teams willing to challenge one another constructively. Cultures that hold steady under pressure. Leaders who run toward the hardest problems because they’re the ones that actually matter.

This Masterclass provides a blueprint for founders navigating AI’s next frontier — one where success is defined not by model novelty, but by the depth and durability of the systems we build.

This post is based on insights from Radical Talks, a podcast from Radical Ventures exploring innovation at the forefront of AI. For more cutting-edge conversations, subscribe wherever you get your podcasts.