Enterprise AI teams are hitting a wall, not because their models can’t reason, but because the underlying workflows were never built for agents. Tasks fail, transfers are interrupted, and the problem worsens as organizations push agents into back-office systems. A new architectural layer is emerging to address this: workflow execution control planes that impose a deterministic structure on the processes that agents are expected to execute.
One of the companies that is bringing this to the forefront is Salesforce, with a new workflow platform that turns administrative workflows into a set of tasks to be completed by specialized agents. Users can upload their processes or use one of the established Blueprints provided by Salesforce, and Agentforce Operations will break it down for agents.
Salesforce SVP of Product Sanjna Parulekar told VentureBeat in an interview that the problem is that many enterprise workflows are not designed for agents. “What we’ve seen with customers is that many times, the flaw in a process is probably in the product requirements document,” Parulekar said. “So when that is loaded into a product, it doesn’t quite work. We can optimize it, remove some things and replace it with an agent.”
Without this dashboard layer, companies could run the risk of deploying agents that increase costs instead of solving their workflow problems.
Make workflow work for agents, not just humans
Companies that deploy agents are learning a costly lesson: Their workflows were designed around gaps in human judgment, not automated execution. Processes that evolved over years of workarounds—vaguely defined steps, implicit decisions, coordination that depends on individuals knowing what to do next—break down when agents are asked to literally follow them.
Even with all of a company’s context at our fingertips, AI systems will struggle to complete tasks if it’s not clear what they’re supposed to do.
Parulekar said his team found that focusing on what makes the process work and breaking it down into more explicit steps and workflows makes 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 because of the session tracking model in the system,” he said.
Parulekar said human checks can be built into the system, making the process 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; the system does it. Unlike more traditional automation tools that direct tasks and agents based on probabilistic decision making, this forces execution into a more predefined deterministic structure.
The problem that presents
Coding a workflow doesn’t fix a broken one. If a process has broken steps, coding it for agents blocks the problem at scale. And once workflows are distributed among agents, the challenge shifts from execution to governance: who owns the process, who validates it, and how it evolves when business conditions change.
Hold teams accountable for taking a hard look at what’s working for them and what’s not.
Organizations should consider that, along with the execution control plane offered by platforms like Agentforce Operations, someone must be responsible for the completion and success of tasks.
Brandon Metcalf, founder and CEO of workforce orchestration company Asymbl, told VentureBeat in a separate interview that the key to both humans and agents following a workflow is a shared goal.
“You have to understand the objective or the agent or the human will not complete the task successfully,” Metcalf said. “Someone has to manage the result that needs to be achieved. It can be a person or an agent.”
The bottleneck has moved. As Metcalf put it, the question is no longer whether agents can reason through a task, but whether the underlying workflow is coherent enough to execute it. For companies that built their processes around human judgment and institutional memory, that’s a more difficult solution than switching to a smarter model.
