Enterprise AI deployments are breaking. Not because the models are bad, but because the workflows they're supposed to follow were never designed with machines in mind.
Salesforce is targeting that gap directly with Agentforce Operations, a new platform that takes back-office workflows and restructures them into discrete tasks that specialized agents can execute. Companies can upload their existing processes or choose from a set of pre-built Blueprints that Salesforce provides. The platform then breaks those workflows down into explicit steps and assigns them to agents accordingly.
The insight driving the product is uncomfortable for a lot of organizations: their processes are broken before any agent touches them.
Sanjna Parulekar, Salesforce's senior vice president of Product, put it plainly. 'What we've observed with customers is that a lot of times, the brokenness in a process is probably in your product requirements document,' she told VentureBeat. 'So when that's uploaded into a product, it doesn't quite work. We can optimize it and cut out some things and replace it with an agent.'
Years of workarounds, informal handoffs, and decisions that lived inside people's heads made existing enterprise workflows functional for humans. Agents don't carry that institutional memory. When asked to follow loosely defined steps literally, they fail. Tasks stall, handoffs break, and costs climb.
Parulekar said the fix starts with making processes explicit. Her team found that breaking workflows into clearly defined steps makes the system more deterministic, which means agents enter the process already knowing their specific role. The platform also builds in session tracing and human checkpoints, giving teams visibility into what's happening and where things go wrong.
The core difference from traditional automation tools is architectural. Most workflow software still relies on agents or systems making probabilistic decisions about what to do next. Agentforce Operations enforces a pre-defined execution path. The system decides the sequence, not the agent.
That structure solves one problem and surfaces another. Encoding a flawed workflow into a deterministic system doesn't fix the flaws. It scales them. Once a bad process is distributed across multiple agents, the failure compounds fast. And when something goes wrong, the question of who owns the process, who validates it, and who updates it when conditions change becomes urgent.
Brandon Metcalf, founder and CEO of workforce orchestration company Asymbl, framed it this way: 'You have to understand the goal or the agent or human won't complete the task successfully. Someone has to manage that outcome that has to be delivered. It can be a person or an agent.'
That accountability gap is the real challenge. Platforms like Agentforce Operations can impose structure, but they can't substitute for someone being responsible for whether the process actually works.
The bottleneck in enterprise AI has shifted. It is no longer about whether models can reason through a problem. It is about whether the workflow underneath them is coherent enough to run. For organizations that built their operations around human judgment and informal knowledge, that is a harder fix than upgrading to a smarter model.
