Workflows in the Age of AI Agents
Agent-based artificial intelligence is considered as the next leap in productivity. The logic is simple viz., assign a task to an agent and let it run. Yet the effectiveness of these systems depends far less on the sophistication of the code and far more on the clarity of the workflow they are given.
A workflow is the structured path through which work moves from start to finish. It defines the sequence of tasks, the hand-offs between people or systems, and the expected outcomes at each stage. Within this path, the process provides the backbone, setting out how each step should be performed, what standards apply, and how exceptions are to be managed. Without these foundations, the very data that agents rely on becomes unreliable. It is like attempting a Six Sigma project without stabilising the process first, because what you are measuring is so inconsistent that any optimisation is built on sand.
This is easy to see in practice. Consider a hospital admission workflow, which involves multiple departments, sensitive data, and critical timing. The journey begins with patient registration, where details are collected, a unique identity is created, and basic documentation such as identification or insurance papers is checked. The next vital step is financial clearance, either through insurance verification or collection of a deposit. If this is skipped or handled late, problems cascade downstream. Once financial clearance is complete, a medical assessment follows, triaging emergencies and routing others for physician review. Bed or room allocation then takes place, coordinated with housekeeping and transport to ensure readiness. Finally, clinical onboarding begins, with nurses briefing the patient, scheduling diagnostics, and updating the treating doctor through the hospital’s systems. From here, care and monitoring continue in a loop of data collection and coordination.
When this workflow is carefully mapped, each step is clear, responsibilities are defined, and the information flow is reliable. Insurance clearance cannot occur after room allocation, because the order is already locked. Registration cannot skip details, because every downstream task depends on them. The mapping creates discipline and ensures that the process holds together.
It is only on this foundation that AI agents add real value. They can automatically verify insurance coverage, trigger alerts when financial clearance is delayed, schedule diagnostic tests in line with physician orders, or update records in real time. But if the underlying workflow is inconsistent for instance, if insurance checks sometimes happen before and sometimes after admission the agent merely accelerates the inconsistency. Instead of solving the problem, it makes it worse.
The same logic applies to outside hospitals. In employee onboarding, for instance, the workflow may run from offer acceptance to first-day orientation, but unless the processes are consistent—laptop provisioning, payroll activation, mandatory training—the onboarding experience becomes fragmented, and any attempt to automate simply multiplies the unevenness.
AI agents are not shortcuts. They are multipliers. They multiply the clarity of a well-mapped workflow and the strength of well-defined processes. But they also multiply the confusion where structure is missing. The real question for organisations is not what agents can do, but whether the workflows and processes in place are clear enough for them to succeed.
The relevance and importance of this becomes even more clear on a ready of the article in McKinsey. They argue that productivity gains do not come from redrawing the structure chart but from rethinking the process itself. Their framework of four levers eliminate, synchronize, streamline, and automate offers a practical way to make workflows more resilient and effective. The traditional disciplines of process mapping and standardisation, sometimes dismissed as dated, are in fact more critical than ever in the age of AI agents. To eliminate is to cut away what is redundant, whether duplicate reports, excessive meetings, or unnecessary checkpoints. To synchronize is to ensure that information flows across boundaries without delay, so that decisions are taken in context and in time. To streamline is to reduce clutter, focusing on what matters most to decision quality instead of drowning people in backward-looking commentary or irrelevant granularity. And to automate is to use digital tools to take over routine work, allowing human judgment and creativity to come to the fore.
Placed against the earlier examples, the relevance is obvious. In hospital admissions, eliminating unnecessary checkpoints, synchronising across clinical and financial functions, streamlining reporting to focus on patient readiness, and automating insurance checks would not only reduce errors but also accelerate outcomes. In employee onboarding, the same four levers would prevent duplication, improve hand-offs, and allow AI agents to truly enhance the experience rather than amplify confusion.
Seen this way, the emphasis on process is not old-fashioned bureaucracy but the very foundation of modern productivity. AI becomes a companion rather than a replacement. It multiplies whatever exists discipline or disorder and the responsibility lies with organisations to ensure that what exists is well designed. Only then can agents elevate performance, reduce wasted effort, and create sustainable value.
