Counting agents is the wrong scoreboard: Better questions for enterprise leaders

Many executives are obsessed with the number of AI agents they run, but this is the wrong metric for measuring AI progress. In workshops with companies across Europe and the Middle East, it became clear that those achieving real value from AI tools didn't start with a focus on agents. Instead, they focused on the specific business problems they wanted to solve and worked backwards from there. The four questions that are steering good companies into Pilot Purgatory are:

1. How many agents should we run?

2. Can AI recommend?

3. Which model tops the leaderboard?

4. Can we find one system that does everything?

Instead, leaders should ask:

1. What outcome are we improving?

2. Can AI finish the job?

3. Does it understand our business?

4. How do the specialists coordinate?

The focus should be on the outcomes AI helps to achieve, rather than the number of agents. Value is shown in real business results, like fewer payroll errors or faster shift scheduling. Agentic applications coordinate teams of specialized agents to deliver business outcomes, while individual agents help with specific tasks. The context of the enterprise, including trusted data and business processes, is crucial for effective AI.

Rather than seeking a single, all-inclusive AI system, organizations should look for coordinated teams of specialists, each knowledgeable in their own domain and working together towards a shared goal.