The AI revolution demands a redesign of work
AI creates organisational value when leaders redesign tasks, decision rights, and accountability around its actual capabilities.
The AI revolution is changing what software can contribute to knowledge work. Leaders should respond by examining how work is organised, which decisions can be supported, and where human responsibility must remain explicit. Access to a capable model is the beginning of that work.
Generative AI can assist with activities such as drafting, summarising, and interpreting information. Systems that combine models with tools can also take actions within a workflow. The move from producing a suggestion to executing a task makes the design of permissions, review, and recovery especially consequential.
Define the work before selecting the agent
Choose a specific outcome and map the tasks required to achieve it. Identify which tasks depend on judgment, which rely on dependable information, and which involve a consequential commitment. This makes it easier to decide where assistance is useful and where approval is necessary.
Consider an illustrative assistant preparing responses to supplier enquiries. It may help find relevant records and draft a reply. A commitment to price, delivery, or a revised order requires clearly assigned authority. Leaders should define that boundary in the workflow and in the system’s permissions.
Measure the complete task
A faster first draft does not automatically mean a more productive process. Include the time spent checking, correcting, coordinating, and handling exceptions. Assess quality on representative work, including cases where the system lacks information or should decline to act.
NIST’s Generative AI Profile addresses risks across the AI lifecycle, including issues such as confabulation and human-AI interaction. Its relevance to leadership is practical: evaluation and risk management need to continue as the system, its users, and its operating context change. (NIST source)
Record the baseline and the required level of performance. A pilot should establish where the tool is dependable, where it needs supervision, and where another approach is more appropriate. Use those findings to set the next level of commitment.
Design responsibility into adoption
Assign an owner for the workflow and identify who maintains the source information, reviews performance, and manages changes. Give employees guidance they can apply to real tasks, including how to identify uncertainty and escalate a problem.
When a system can act, limit access to the actions and information required for its purpose. Make consequential activity traceable and define how people can interrupt or recover the process. The control should reflect the impact of the action.
Turn local gains into an organisational capability
The wider opportunity lies in redesigning the work around demonstrated strengths and known limits. That can include changing handoffs, improving source information, reducing repeated effort, and giving people more time for the judgment the task requires.
Leaders should develop this capability deliberately. AI becomes part of the organisation’s performance through the relationship between technology, people, information, and authority. That relationship deserves as much attention as the model itself.
A new essay connecting my manufacturing and operations perspective with enterprise architecture. Examples are illustrative unless attributed to a named source.