Enterprise AI teams are hitting a wall—not because their models can’t think, but because the workflows beneath them were never built for agents. Tasks fail, handovers are disrupted, and the problem is compounded as organizations push agents deeper into back-office systems. To solve it, a new layer of architecture is emerging: workflow execution control planes are expected to be implemented, which apply a deterministic structure to process agents.
One company bringing this to the fore is Salesforce, with its new workflow platform that turns back-office workflows into a set of tasks for specialized agents to complete. Users can upload their own processes or use one of the specific plans provided by Salesforce, and Agentforce Operations will break it down for agents.
Sanjna Parulekar, senior vice president of Product at Salesforce, told VentureBeat that the problem is that many enterprise workflows are not built for agents. “What we’ve seen with customers is that many times the process disruptions are in your product requirements document,” says Parulekar. “So when it’s loaded into the product, it’s not fully functional. We can optimize it and cut some things out and replace it with an agent.”
Without this dashboard layer, businesses risk deploying agents that drive up costs instead of solving workflow problems.
Making workflow work for agents, not just humans
Enterprises deploying agents are learning an expensive lesson: Their workflows are designed around human judgment gaps, not machine execution. Processes that have evolved over years of solution—loosely defined steps, implicit decisions, coordination that depends on individuals knowing what to do next—break down when agents are literally asked to follow them.
Even with the entire context of an enterprise at your fingertips, AI systems will struggle to complete tasks if they aren’t clear on what to do.
Parulekar said his team focused on what caused the process, and breaking it down into clearer steps and workflows made the system more deterministic. Then, when platforms like Agentforce Operations introduce agents, those agents already know their specific tasks.
“It forces companies to rethink their processes and introduces observability into the mix due to the session tracking model in the system,” he said.
Parulekar said human checks can be fed into the system, so the process is more transparent.
What sets this approach apart from other workflow automation offerings is that it doesn’t rely on agents to decide what to do next; system does. Unlike more traditional automation tools, which direct tasks and agents to make probabilistic decisions, it enables execution in a more predetermined, deterministic structure.
The problem it presents
Coding a workflow doesn’t fix what’s broken. If there are defective steps in the process, the coding problem for agents locks in scale. And once workflows are distributed among agents, the problem shifts from execution to management: who owns the process, who approves it, and how it evolves as business conditions change.
This puts the onus on teams to take a hard look at what works and what doesn’t.
Organizations must consider that in addition to the execution control plane offered by platforms such as Agentforce Operations, someone needs to be held accountable for task completion and success.
Brandon Metcalf, founder and CEO of workforce orchestration company Asymbl, told VentureBeat in a separate interview that the key to following workflow is a shared goal for both people and agents.
“You have to understand the goal, or the agent or human will not complete the task successfully,” Metcalf said. “Someone must manage the result to be delivered. It can be a person or an agent.”
The bottleneck was removed. As Metcalf says, the question is not whether agents can perform a task, but whether the workflow beneath them is consistent enough to perform. For businesses that base their processes on human judgment and institutional memory, this is a more difficult solution than changing to a smarter model.
