The wall enterprise AI teams keep hitting
Many teams have had the same experience with AI agents. The model handles a task well in a demo. Then it goes into production, and things start to slip. Tasks fail, handoffs break, and the problems grow as agents take on more back office work.
A VentureBeat article by Emilia David argues that the cause is usually not the model's ability to reason. It is that the workflows underneath the agents were never built for them. The article reports on Salesforce's launch of Agentforce Operations, and on a broader idea it describes as an emerging architectural layer: workflow execution control planes that impose a deterministic structure on the processes agents are expected to run.
This post covers what the article reports about the launch, the larger lesson behind it, and how any team can apply that lesson.
What Salesforce announced
According to the article, Agentforce Operations is a workflow platform that turns back office workflows into a set of tasks for specialized agents to complete. Users can upload their own processes or start from Blueprints provided by Salesforce, and the platform breaks the process down for agents.
Sanjna Parulekar, Salesforce's senior vice president of product, told VentureBeat that many enterprise workflows are simply not built for agents. "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 said. "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."
The article highlights a few other points from the interview:
- Breaking a process into more explicit steps makes the system more deterministic, so agents already know their specific tasks when they are introduced.
- The system includes session tracing, which Parulekar said adds observability.
- Human checks can be built into the process to make it more transparent.
- The system, rather than the agent, decides what happens next, which the article presents as the main difference from tools that route work based on probabilistic decisions.
The deeper lesson: workflows built for people break for agents
The most useful part of the article applies well beyond any one product. It explains that many enterprise processes were designed around human judgment. Over years of workarounds they picked up loosely defined steps, decisions nobody wrote down, and coordination that depends on someone simply knowing what to do next.
People handle that ambiguity without noticing. Agents do not. When an agent is asked to follow a process literally, the gaps show up as failures. As the article puts it, even with all of an enterprise's context available, an AI system will struggle if it is not clear what it is supposed to do.
The catch: codifying a broken process locks it in
The article is honest about the limits of this approach. Writing a workflow down does not fix it. If a process has flawed steps, encoding it for agents can lock in the problem at scale. And once work is spread across several agents, the challenge shifts from execution to governance: who owns the process, who validates it, and how it changes when the business changes.
Brandon Metcalf, founder and CEO of workforce orchestration company Asymbl, told VentureBeat that the key to both humans and agents following a workflow is a shared goal. "You have to understand the goal or the agent or human won't complete the task successfully," he said. "Someone has to manage that outcome that has to be delivered. It can be a person or an agent."
The article's closing point is that the bottleneck has moved. The question is no longer whether agents can reason through a task. It is whether the workflow underneath them is coherent enough to run.
How to make a workflow ready for agents
You can apply this lesson with any tool. Before putting an agent on a process, work through a few steps.
- Write down the real process. Include the informal steps people actually take, not just the official version.
- Fix it before you encode it. Remove steps that exist only as workarounds. Automating a broken process only makes it fail faster.
- Make each step explicit. Define the input, the output and what done looks like for every step.
- Decide who owns each step. A fixed rule, an agent or a person. Ambiguity here is where handoffs break.
- Add checks before sensitive actions. Put a person in front of anything that spends money, contacts a customer or changes an important record.
- Name an owner for the outcome. Someone has to be accountable for the whole process, not just its parts.
- Make every run visible. You should be able to see where a run stalled without reconstructing it from logs.
A quick example
Take a common back office process: handling an incoming vendor invoice. The informal version might be "someone checks it and gets it paid." An agent ready version is explicit. The invoice arrives and is classified. Key fields are extracted. The amount is matched against the purchase record. Matches under an agreed limit move to the next step automatically. Anything over the limit, or anything that does not match, goes to a named approver with a short summary. Once approved, the record is updated and the vendor is notified.
Each step has a clear owner, and the agent only makes judgment calls where judgment is actually needed.
Where Innflow fits
Innflow is the all in one workspace for AI agents, and it is designed for this kind of structure. You can define your processes and share clear instructions so agents know how you want the work handled. Workflows let you choose conditions and assign the next step to a teammate or an AI agent, so the predictable parts of a process stay predictable. You can choose where approval is needed, coordinate actions across connected tools from one workflow, and follow each agent's progress from one workspace.
That lets you start with the steps you trust, add agents where they help, and keep a person accountable for the outcome.
Fix the workflow first
The article's message is a useful reset for anyone deploying agents. A smarter model will not rescue an unclear process. Make the workflow explicit, fix it before you encode it, and give every step and every outcome an owner. If you want one workspace to build agent ready workflows, you can try Innflow at innflow.ai.
Frequently asked questions about Agentforce Operations
What is Agentforce Operations?
According to the source article, it is a Salesforce workflow platform that turns back office workflows into tasks for specialized agents to complete.
Why do AI agents fail in production?
Often because the workflow was built for people. Informal steps, missing handoffs and unclear rules that people work around trip agents up.
How do I make a workflow ready for agents?
Write the steps down, define inputs and outputs, mark the decisions that need a person, and fix the process before automating it.
Related reading
- Agentic Orchestration: What It Means for Enterprise AI
- How AI Agents Are Automating Vehicle Support
- How to Use AI for Business Automation Without a Tech Team
Source: Emilia David, VentureBeat, "Salesforce launches Agentforce Operations to fix the workflows breaking enterprise AI" (May 1, 2026). Read the original.












