The agent value in process management

Last month we explored the human value in context management, this month, we turn to what runs on top of that context: the process layer. 

Earlier this month, the industry drew a line it had been approaching for some time. The two leading open protocols for AI agents, one governing how agents access knowledge and tools, the other governing how agents communicate and coordinate with each other, were brought together under a single foundation with over 250 member organisations. What the agent knows is one discipline, what the agent does is another. 

That separation matters commercially. BCG's latest research on agentic enterprise operations found that organisations redesigning end-to-end processes around agents are seeing threefold productivity gains and significantly reduced cycle times, while those bolting agents onto existing workflows see returns of just ten to twenty percent. Yet Forrester reports that fewer than fifteen percent of firms will activate the agentic capabilities already sitting inside their enterprise software this year. The gap is not in the technology but rather in the process design surrounding it. 

This is where we see the real opportunity, and where the agent value in process management amplifies the human value in everyday work. 

Redesign the process, don't decorate it 

The instinct is to drop an agent into a workflow that already exists and call it automation. We'd propose a different starting point: what should this process look like when an agent handles the routine, and a human directs the exceptions? That requires designing the workflow around what each does best, not speeding up what was already there. 

Separate what the agent knows from what the agent does 

We’ve made the argument that structured, deliberately engineered context is the foundation. August's extension: keep the process layer; the orchestration, the decision logic, the approval gates; distinct from the knowledge it draws on. When the two blur together, you get agents that sound confident but behave inconsistently, because no one can see where the knowledge ends and the automation begins. Clarity here is what makes agents governable and their outputs trustworthy. 

Design the human role into the process, not around it 

The most effective agentic processes are not ones where humans are removed and added back as a safety net. They are ones where the human contribution; judgement, creativity, relationship; is designed into the workflow at the points where it has the highest impact. Get the process layer right and you are not replacing human value; you are building the conditions for it to compound.  

This is the third in our short series on what the move from content to context asks of a business. Next month, we look at the interface: how agents, customers and your brand meet.