Salesforce Unleashes Agentforce Operations: A Deterministic Control Plane to Fix the Breaking Workflows That Stymie Enterprise AI

Salesforce today launched Agentforce Operations, a workflow execution control plane designed to impose deterministic structure on the chaotic, human-centric processes that are causing enterprise AI initiatives to fail. The new platform turns back-office workflows into explicit, machine-readable tasks that specialized agents can reliably execute, addressing a critical failure point where agents break due to vague or implicit steps.

“The brokenness in a process is probably in your product requirements document,” said Sanjna Parulekar, senior vice president of Product at Salesforce, in an interview. “When that’s uploaded into a product, it doesn’t quite work. We can optimize it, cut out some things, and replace it with an agent.” Without this control layer, enterprises risk deploying agents that increase cost rather than fix workflow problems.

The Core Problem: Human‑Centric Workflows Break Agents

Enterprise AI teams are hitting a wall — not because their models can’t reason, but because the workflows underneath them were never built for agents. Tasks fail, handoffs break, and the problem compounds as organizations push agents deeper into back‑office systems. Processes that evolved through years of workarounds — loosely defined steps, implicit decisions, and reliance on human judgment — collapse when agents follow them literally.

Even with full enterprise context, AI systems struggle if instructions are not explicit. Parulekar explained that focusing on what makes a process tick and breaking it into explicit steps makes the system more deterministic. “It forces companies to rethink their processes and introduces observability into the mix because of the session tracing model in the system,” she added.

The Solution: A Deterministic Workflow Execution Control Plane

Agentforce Operations provides a control plane layer that enforces a pre‑defined, deterministic structure on the workflows agents are expected to run. Users can upload their own processes or select from Salesforce’s set of Blueprints. The platform then decomposes the workflow into discrete tasks for specialized agents, routing work based on system directives — not probabilistic AI decisions.

This differs sharply from traditional automation tools, which often rely on agents making probabilistic decisions about what to do next. Instead, Salesforce’s approach imposes deterministic execution, ensuring that agents only perform the exact tasks assigned to them. Parulekar noted that human checks can be built into the system, making the process more transparent.

Background: Why Enterprises Are Struggling

Enterprises deploying agents are learning a costly lesson: Their workflows were designed around human judgment gaps, not machine execution. Over time, processes accumulate workarounds that require tacit knowledge — something AI lacks. When agents are deployed without re‑engineering these workflows, they either fail or produce unpredictable results, undermining the business case for AI.

Salesforce’s new product directly addresses this by forcing companies to codify their processes upfront. The system’s session tracing model provides observability into every step, allowing teams to see exactly where agents succeed or break. This, Parulekar said, “makes the process more transparent” and enables continuous improvement.

What This Means

Agentforce Operations signals a strategic shift: Workflow design must be rethought for agents. Enterprises that adopt this control plane will likely see more reliable AI agents, reduced cost overruns, and faster scaling of automation. However, the product also introduces a new risk: codifying a flawed workflow locks in problems at scale. If a process has broken steps, encoding it for agents amplifies those failures.

“Without this control panel layer, enterprises could risk deploying agents that increase cost rather than fix their workflow problems,” Parulekar warned. The platform’s success will depend on how well organizations use its observability features to identify and fix underlying process flaws — not just automate them. For now, Salesforce has thrown down a gauntlet to competitors: tame the workflow chaos or watch enterprise AI stall.

— Breaking news coverage by VentureBeat