Football fans will remember the huge excitement that accompanied Real Madrid’s Galácticos policy. It was electrically tangible the first time around, from 2000 to 2006. We all felt it again during its later second wave, when club president Florentino Pérez Rodríguez was back. But the reboot felt different. His vision was reinstated, but we were no longer blind adherents.

The Galácticos policy was built on the personal Pérez philosophy that if a football club was able to assemble the world’s best individual talents, it would automatically produce a winning team.

With a massive €480 million cash injection from the sale of the club’s Madrid training ground, Ciudad Deportiva, Pérez set about buying the world’s best players. Over the course of the policy the roster had included Alonso, Beckham, Figo, Kaká, Modrić, Owen, Ramos, Raúl, Robino, Rodríguez, Ronaldo, and of course the mighty Zidane.

Any group of these players would form a lineup of All Stars more impressive than any ever-before assembled team, excepting only earlier incarnations of the Real Madrid Galácticos. With all that player talent, the club couldn’t help but win.

But often it didn’t. The plan, it turned out, had a fatal flaw. In short, superstars’ individual brilliance isn’t the only ingredient for success in a team sport. It cannot compensate for the absence of system, structure, and clearly defined roles for everyone on the team.

The Galácticos parallel to agentic AI is almost exact.

Signing the best agent ≠ winning the workflow. Just as buying Beckham didn’t solve Real Madrid’s defensive midfield problem, deploying a state-of-the-art AI agent doesn’t fix a broken or poorly designed process. As McKinsey rightly points out in their article One year of agentic AI, it’s about the workflow, not the agent.

An absence of clear roles = chaos. The Galácticos famously struggled because everyone was a superstar. No one wanted to do the less glamorous pressing, tracking back, providing defensive cover. In agentic AI, the same trap exists. Companies often deploy AI agents willy-nilly, before defining the outputs that to which each is accountable, where human judgment must be injected, and who has ownership of the edge cases. The result is the same: beautiful in training, dysfunctional in the match.

The “AI slop” problem = the Galácticos problem. Whilst the Galácticos policy was impressive in the Bernabéu showcase, it was very often frustrating when it actually mattered. McKinsey says that agents need onboarding, job descriptions, and feedback loops, in just the same way that even superstars need a coach and a system in which to play the game as a team.

Player versus team. Zidane, one of the earliest Galácticos players, returned to Real Madrid in the waning days of the policy. This time, though, he came in as manager. He and the team won three consecutive Champions League titles by fixing the fatal flaw in the system. He built a team system around the stars.

It’s the neat second layer that’s essential to making Agentic AI successful. The same technology (or talent) produces dramatically different results when it is deployed within a deliberate structure.

Our sector is having a Galácticos moment with Agentic AI. We are all excited about recruiting the best AI models and tools. But we are underinvesting in the dull tactical setup – the workflow design, the human-agent collaboration, the evaluation frameworks – that will determine whether our underwriting teams actually win.

Usually my Insurance Technology Diary entries about AI end with the caveat that they are accurate only to the best of my knowledge at the time of writing, due to the rapid pace of AI progress. Not this one.

The lesson from Bernabéu will always be true. It’s the system, not the stars, that lifts the trophy.

Guillaume Bonnissent is CEO of Quotech.