I recently had a discussion with a client about AI agents and how they could reshape enterprise operations.
During our conversation, I shared an interesting perspective from friends working in the automotive industry. They had been telling me about the rise of Dark Factories in Asia, where manufacturing facilities are becoming increasingly autonomous, with machines coordinating production, controlling quality and managing operations with very limited human intervention.
What caught my attention wasn't the technology itself, but the industrial thinking behind it.
The automotive industry didn't achieve operational excellence simply by introducing robots. The real transformation came from redesigning entire production systems around automation, standardization and continuous improvement.
I see a strong parallel with enterprise AI today.
Over the past two years, companies have invested heavily in AI agents, often approaching them as individual use cases. Each team identifies opportunities, experiments with technologies and develops its own solutions.
This makes sense in the early stages, but building an agent that can execute a task is very different from operating hundreds of agents reliably across an organization.
And more importantly, automating tasks doesn't necessarily translate into measurable financial benefits.
As I often say:
"Without rethinking the operating model, AI agents won't deliver cost savings. They'll simply create more time for coffee and cigarette breaks."
Behind this provocative statement is a serious point. Productivity gains don't automatically translate into P&L impact. Unless organizations rethink how work is structured, how responsibilities are distributed and how processes are executed, much of the expected value may never materialize.
This is where the Dark Factory concept becomes particularly interesting.
Imagine an enterprise where agents are not deployed as isolated applications but operate within a common production environment. Some agents plan activities, others execute them, while others validate results, monitor performance or manage exceptions.
The objective is no longer to automate individual tasks but to redesign and industrialize entire business processes.
This shift has significant implications for enterprise architecture, but also for operating models.
Who is accountable when several agents contribute to the same business outcome? How do we manage dependencies between agents? How do we ensure consistency, traceability and quality? And how do we determine which decisions should remain under human control?
These questions cannot be answered simply by selecting a better LLM or adopting another agent framework.
They require the same operational discipline that manufacturing has developed over decades.
There is also an interesting parallel with DevOps. Software engineering became more scalable when organizations standardized development, testing, deployment and monitoring. AgentOps could play a similar role for enterprise AI, although autonomous agents introduce additional challenges around reliability, governance and decision-making.
But beyond technology, I believe the most significant transformation will be organizational.
Most business processes have been designed around people, with technology introduced to support or automate specific activities.
AI agents create an opportunity to reconsider that assumption and redesign certain processes around autonomous execution, with human intervention focused on supervision, exceptions and decisions requiring judgment.
This requires a different conversation with business leaders. Less about the number of agents deployed, and more about process performance, accountability, operating models and the economics of automation.
Looking back at my conversation with the client, I think the automotive industry offers an important lesson.
Manufacturers didn't become more efficient simply because robots became more capable. They became more efficient because they learned how to organize entire production systems around them.
We are still spending considerable effort making individual AI agents more capable. The harder question, and probably the more valuable one, is how to make them work together as a dependable part of the business.
The next competitive advantage in enterprise AI may not come from building smarter agents, but from learning how to operate them together.