AI workflow redesign fails at scale when organizations stop at the task level and never reach the subtask. The gap between knowing which roles are AI-ready and actually rebuilding how work runs is where most enterprise AI programs stall. Sarah Bernstein, VP of Organization Transformation and Capability Development at Lumen Technologies, is working through that gap in real time, transforming a legacy telco into a digital network service provider across every function and workflow simultaneously.
This conversation, from Reejig's Work Blueprint series, covers why subtask depth is non-negotiable for true AI reimagination, why just-in-time learning is the missing piece after workflow redesign, what it means to make people pilots rather than passengers, and why AI fear-mongering by AI companies made the adoption challenge harder for everyone.
Task-level data tells you where AI potential exists. Subtask-level data tells you how to actually rebuild the work. These are different problems, and confusing them is why so many AI programs produce insight but no redesign.
The practical implication is significant. When Lumen mapped invoice processing, the agent does not take the full payment invoice. The subtasks, each step required to complete the payment, are where the agent operates. Without that subtask map, the agent build is incomplete. Without the work context at each handoff point, which system the work moves to and under what conditions, the agent Blueprint is only 80 percent right. As Siobhan Savage, Founder and CEO of Reejig, observed: "80% is not good enough if we want to be able to build the capability across all of our teams to do this at scale."
Every enterprise is deploying AI. Almost none can see the work they're deploying it into. The Work Context Graph is the structured record that makes that visibility possible: Work Architecture as the foundation, Work Context as the reality of how work actually flows, and Work Record as the governance layer. That is the infrastructure that makes agent building reliable at enterprise grade.
Most enterprises have solved the workflow redesign problem and are now discovering the adoption problem. Agents are built. Workflows are redesigned. Employees receive prompt training. Then nothing changes, because no one showed people their actual new way of working.
Bernstein's response at Lumen was to build learning journeys directly from the agent design session. Out of the workflow redesign workshop, two parallel tracks ran: one building the learning path for the new agent-enabled workflows, one building the always-on judgment and adaptability capability that every function will need as the redesign continues. The learning is built into the redesign, not added afterward.
Reejig's Work Operating System now surfaces standard operating procedures automatically when a workflow is published, giving employees a concrete, step-by-step picture of the new way of working. That is Stealth Change Management in practice: continuous, embedded change delivered inside the systems your people already use. No kickoff meetings. No separate program. Just the work, updated, with the new workflow shown clearly to every person it affects.
People do not naturally embrace change, and the AI era has added a specific layer of fear that makes adoption harder. Bernstein's framing for Lumen's second-day session was clear: the goal is to move people from passengers, going along for the journey, to pilots, actively applying judgment, curiosity, and product thinking to what their function is becoming.
The clarity piece is the practical obligation. Employees who do not know which tasks are changing, which are staying human, and what they should be doing with freed-up capacity will not redirect that capacity to higher-value work. The communication has to reach the individual role level, not just the function or the business unit. This is the activation gap between a redesigned workflow and an organization that actually works differently.
The AI companies that built the narrative of job elimination created a direct obstacle for every enterprise trying to drive genuine AI adoption. When employees are primed to see AI as a threat, they resist the new workflows that would actually make them more capable. The harder the fear-mongering, the harder the change management.
The reality inside Lumen's transformation is the opposite of that narrative. Functions are merging rather than disappearing. Design, engineering, and product management are converging as AI raises the subject matter expertise floor. Early career people are being accelerated by AI wrapping, not replaced by it. The opportunity is for more people to operate at a higher level, sooner, not fewer people doing less.
The entry-level job crisis narrative misses the more important question: how do you design AI into an early career person from day one? Bernstein's observation from working with early career professionals at Lumen is that the trajectory question matters more than the role question.
If agents handle the repetitive execution that used to take three to four years to get through, and the early career person is building judgment, adaptability, and domain understanding from the start, the question becomes whether a two-year journey to genuine capability is achievable where a four-year journey was the norm.
The second dimension is capturing the expertise of twenty-year domain veterans and pairing it with the AI-first instincts of people just entering the workforce. Neither group alone has what the other has. The organizations that figure out how to combine institutional expertise with AI-native thinking, and wrap agents around both, are the ones that compound fastest.
|
CHRO Focus |
CIO Focus |
Shared Outcome |
|
Subtask-level work mapping and AI-led versus human-led classification at workflow depth |
Agent inventory, approved AI stack, and Work Context Graph infrastructure |
Agent Blueprints that are complete, context-rich, and ready to deploy at enterprise grade |
|
Just-in-time learning journey design, built from the redesigned workflow |
Standard operating procedure generation and workflow publication tooling |
Employees who know their new way of working from day one, not after a separate training program |
|
Pilots-not-passengers change leadership: equipping people leaders to guide their teams through continuous redesign |
Change logging and audit trail for every workflow and agent modification |
An organization that absorbs continuous AI-driven change as standard operating rhythm |
Why is subtask-level data more important than task-level data for AI workflow redesign? Task-level data identifies where AI potential exists in a role. Subtask-level data is what makes actual workflow reimagination possible. Agents do not take over entire tasks. They operate on specific subtasks, at specific handoff points, with specific system context. Without that depth, agent builds are incomplete, and the redesign stays theoretical rather than operational. Going to subtask depth takes more time upfront and removes ambiguity from every deployment decision that follows.
What is just-in-time learning and why does it matter for AI adoption? Just-in-time learning is the practice of building capability development directly into the workflow redesign process, so employees know their new way of working from day one of deployment. Most enterprises deploy agents and issue prompt training. Employees then fill freed-up time with whatever they were already doing, because no one showed them concretely what to do instead. Just-in-time learning closes the gap between a redesigned workflow and an organization that actually operates differently.
What is the Work Context Graph and why is it the foundation for agent building? The Work Context Graph is the structured record of how a company runs, unifying Work Architecture, Work Context, and Work Record in three layers. Work Architecture maps every role, workflow, task, and subtask. Work Context captures the handoffs, dependencies, sequences, and policies that govern how work actually flows. Work Record is the governance audit trail. Together, these layers give agents the context they need to operate reliably at enterprise grade, rather than producing outputs that are 80 percent right.
What does it mean to make people pilots rather than passengers in AI transformation? Pilots apply judgment, curiosity, and product thinking to what their function is becoming. Passengers go along for the journey without understanding what is changing or why. Making people pilots requires giving them explicit clarity: which tasks are changing, which are staying human, what the new workflow looks like, and what they should be doing with the capacity they gain. Without that clarity, even well-designed AI deployments fail to change how people actually work.
How should organizations think about early career people in the AI era? Early career people should be accelerated by AI, not replaced by it. If agents handle the repetitive execution that once took years to get through, early career people can build judgment, adaptability, and domain understanding from the start, potentially moving through what was a four-year development trajectory in two. The organizations that redesign entry-level roles to wrap AI around people from day one, rather than eliminating those roles, protect their expertise pipeline while gaining AI-native capability faster.
Why did AI fear-mongering make enterprise adoption harder? AI companies that built a public narrative of job elimination primed employees to see AI as a threat before enterprises began their real redesign work. Employees who expect AI to eliminate their roles resist the new workflows that would actually make them more capable. The harder the fear-mongering, the more resistance enterprises face in driving genuine adoption, and the more work is required to establish the psychological safety and clarity that change readiness depends on.
The real work of AI transformation happens after the workflow is redesigned. Showing every person affected by that redesign exactly how to work in the new way, building learning into the deployment rather than bolting it on afterward, is what converts agent investment into actual change in how the organization operates. The enterprises that get this right are the ones treating AI redesign and people enablement as the same program, not sequential ones.
Book a demo to see how Reejig's Work Operating System maps your work at the subtask level, generates Agent Blueprints with full work context, and surfaces just-in-time learning journeys for every redesigned workflow.