How Leaders Should Redesign Work Around AI Agents
Adding agents to old workflows leaves much of their value on the table. Leaders need to rethink who does what, where decisions sit and how quality is controlled.
By Sophie Keller, Women in AI Editorial Fellow ยท 28 August 2026
The easiest way to introduce AI into a company is to add it to an existing task. The harder and more valuable step is to redesign the work itself.
Microsoft's 2026 Work Trend Index argues that leaders increasingly need to decide what humans do, what agents do and how the two collaborate. That is an operating-model question, not a software rollout.
Start with the workflow, not the tool
Take a process such as customer onboarding, procurement or research.
Map the steps, decisions, handoffs and bottlenecks. Then ask which parts are repetitive, which require judgement and which depend on information scattered across systems.
Only after that should the team decide where agents belong.
Separate execution from judgement
Agents can be strong at gathering information, drafting, routing, checking and executing structured multi-step tasks.
Humans remain valuable where objectives are ambiguous, trade-offs matter, accountability is important or the consequences of error are high.
The goal is not to automate the largest possible percentage of the workflow. It is to allocate work intelligently.
Redesign controls at the same time
If an agent takes over part of a process, the old control environment may no longer make sense.
A manager who previously reviewed every step may need to move toward sampling, exception handling and monitoring. New controls may be required around tool permissions, audit logs and escalation.
Give employees authority, not only automation
One of the more interesting findings in Microsoft's 2026 research is the idea that agents can increase human agency when organisations are designed to support it.
If people spend less time on repetitive execution, they can spend more time directing work, making decisions and improving the process.
But that only happens if managers actually redesign roles. Otherwise employees simply receive more output to supervise.
Measure the redesigned workflow
Do not compare an agent against one isolated task if the whole process has changed.
Measure end-to-end cycle time, quality, customer outcome, cost and employee workload.
A local improvement can create a downstream bottleneck, so the system-level view matters.
Leadership is the constraint
The technical team can build an agent. It cannot decide the future operating model of the company by itself.
That requires leaders who understand the work well enough to redesign responsibilities and who are willing to change structures that existed before AI.
The Women in AI Global Summit will explore this intersection between AI capability and organisational change with practitioners and enterprise leaders in London. The most useful examples are likely to be the ones that show not only what the technology can do, but how companies changed the work around it.
The next phase is organisational
The first phase of generative AI adoption was about access. The next phase is about redesign.
Companies that simply attach agents to old workflows may gain incremental efficiency. Companies that rethink the workflow itself have a chance to create something more substantial.