One quiet effect of AI at work is that more people are discovering how many of their processes were never really processes. They were habits, shortcuts, exceptions, social agreements, and small acts of interpretation that worked because a human knew when to bend the rule. Then somebody tries to automate the work and the convenient story collapses. The system does not understand "usually", "if it feels risky", or "ask someone with context" unless those phrases are turned into decisions.
Vague does not always mean lazy
Some ambiguity is just weak thinking with a polite name. A team has not decided who owns the decision, what quality means, or which tradeoff matters, so the work stays vague because vagueness protects everyone from exposure. This is not flexibility. It is avoidance.
But there is another kind of ambiguity that is more useful. It exists because the world is not stable enough to deserve a rigid rule. A customer request may be technically possible and strategically wrong. A sales promise may be attractive and dangerous. A product exception may be worth making once, but not worth turning into a standard. In those cases, ambiguity is not the absence of thinking. It is a space where judgment can still enter.
Automation has trouble with that distinction. It asks for instructions before the team has agreed what the work really means.
Automation exposes the negotiation
When a company tries to automate a vague workflow, the conversation often looks technical on the surface. Which model should we use? Where should the agent sit? How do we connect the tools? These questions matter, but they are rarely the first problem.
The harder questions are less glamorous. Who is allowed to approve an exception? What level of confidence is enough? Which mistakes are acceptable because they are cheap, and which are unacceptable because they damage trust? When should the system stop and ask a human? Who carries the consequence when the automated decision is formally correct but commercially stupid?
This is why AI projects sometimes feel like they are blocked by implementation when they are really blocked by unresolved negotiation. The vague part of the process was not empty. It was holding disagreement, status, risk, and sometimes a quiet lack of courage.
Clean workflows can hide dirty work
There is a specific kind of diagram that makes organizations feel better about themselves. The boxes are clear, the arrows move in one direction, and every stage has a name. It looks like a process. Often it is only the official version of the process.
The real work happens in the exceptions. Someone knows that this client needs a different tone. Someone remembers that this metric is misleading in Q4. Someone understands that a request should be delayed, not rejected, because the relationship matters. None of this looks elegant in a workflow diagram, but it may be the difference between competence and mechanical compliance.
If automation only captures the official process, the organization may become faster at the least interesting part of the work. The center gets smoother. The edges become more expensive. And because the output looks cleaner, it can take longer to notice that the judgment has moved somewhere else.
The useful question is not whether it can be automated
Technically, more work can be automated than many people assumed a few years ago. That is real. The more important question is whether the work has been understood well enough to deserve automation.
This means asking what the vague part is doing. Is it hiding poor ownership? Then automation may force a useful decision. Is it compensating for a broken system that nobody wants to confront? Then automation may only make the broken system faster. Is it protecting necessary discretion in a relationship, a product judgment, or a risk decision? Then removing it may create precision that looks professional and behaves badly.
The difference is subtle, but it matters. A process that is unclear because people avoid responsibility is not the same as a process that is open because reality requires interpretation.
Precision has a cost
Better teams do not worship clarity as a slogan. They treat it as a cost-benefit decision. They clarify thresholds where repeatability improves the work. They define escalation rules where risk changes. They document intent, not only steps. They leave room for human judgment where the situation is rare, relational, or expensive to misread.
This is less exciting than saying that AI will automate entire functions. It is also more honest. Many functions are not bundles of tasks waiting to be compressed. They are layers of decision-making, trust, context, and social risk. Some of those layers can be encoded. Some need to be named before they can be improved. Some should probably remain human for longer than the automation deck suggests.
The uncomfortable part is that AI does not only automate work. It asks organizations to say what they mean. Sometimes that is useful because it removes laziness disguised as flexibility. Sometimes it is dangerous because it removes the human discretion that made the work sane. The difference will not be obvious from a demo. It appears when something unusual happens and the system has to choose. At that point the question is no longer whether the process is automated. The question is whether the team understood the process before giving it to a machine. That is less impressive, but probably more honest.