Salesforce introduces Agentforce Operations to streamline workflows that disrupt enterprise AI

Enterprise AI teams hit a wall – not because their models can’t think, but because the workflows underneath them weren’t designed for agents. Jobs fail, handoffs break down, and problems compound as organizations push agents deeper into back-office systems. A new layer of architecture is emerging to address it: workflow control planes that impose a defined structure on the processes agents are expected to run.

One of the companies bringing this forward is Salesforce, which has a new workflow platform that turns back-end workflows into a set of tasks for specialized agents to complete. Users can upload their own processes or use one of the Salesforce-provided Blueprints sets, and Agentforce Operations will roll it out to agents.

Salesforce’s senior vice president of Product, Sanjna Parulekar, told VentureBeat in an interview that the problem is that many business workflows are not designed for agents. “What we’ve seen from customers is that most of the time, process breakdowns happen in your product requirements document,” says Parulekar. “So when that’s loaded into the product, it doesn’t work as well. We can optimize it and cut out other things and replace them with an agent.”

Without this layer of control panel, businesses can run the risk of deploying agents that drive up costs rather than fix their workflow problems.

Making workflow work for agents, not just people

Businesses deploying agents are learning an expensive lesson: Their workflows were designed around human judgment gaps, not machine execution. Processes that have evolved over the years of work methods – loosely defined steps, clear decisions, communication that depends on individuals what to do next – break when agents are asked to actually follow them.

Even with the entire business context at its fingertips, AI systems will have difficulty completing tasks if it is not clear what they should be doing.

Parulekar said his team found that focusing on what makes a process tick and breaking it down into clear steps and workflows makes the process more efficient. Then, when platforms like Agentforce Operations introduce agents, those agents already know their specific tasks.

“It’s forcing companies to rethink their processes and introduce hybrid analytics because of the session tracking model into the system,” he said.

Parulekar said that people’s checks can be built into the system, so the process is transparent.

What sets this approach apart from other automated workflow systems is that it does not rely on agents to decide what to do next; the program does. Unlike many traditional automation tools that move tasks and agents into potential decision-making, this forces operations into a pre-defined, predetermined structure.

The problem that presents

Fixing a workflow doesn’t fix a broken one. If a process has faulty steps, coding it into agents locks the problem to scale. And when workflows are distributed across agents, the challenge shifts from implementation to control: who owns the process, who validates it, and how it changes when business conditions change.

It puts the onus on teams to take a hard look at what works for them and what doesn’t.

Organizations must consider that, along with the operational control plane provided by platforms such as Agentforce Operations, someone must 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 both people and agents following workflows is a shared goal.

“You have to understand the purpose or the agent or person will not complete the job successfully,” said Metcalf. “A person must be in charge of that result to be delivered. It can be a person or an agent.”

The bottle is gone. As Metcalf puts it, the question is no longer whether agents can think about work, but whether the workflows underlying them are coherent enough to do it. For companies that have built their processes around human judgment and institutional memory, that’s a tougher adjustment than swapping for a smarter model.