Most organizations misunderstand the AI transformation journey. They imagine it as a linear progression from experimentation to scale, or from one maturity level to the next. But the real journey is much less tidy, following a more ‘squiggly’ path. It does not move cleanly from human-led operations to AI-led operations. It passes through a long, difficult middle phase in which humans and AI must operate together and where neither humans nor AI are really in charge. We called that phase Hands Ready after the experience we’ve all had with our cars on their own journey to autonomy.
That autonomous journey involves three phases. Hands On, where the car provides supportive features like braking or lane alerts, but the human remains fully in control. Hands Ready, where the car drives autonomously in most cases but the human must remain vigilant and ready to intervene instantly, and where control and responsibility are therefore shared. Hands Off, where the car manages operational control while the human’s job is obviously to determine the desired destination and journey objectives (stops along the way to pick up family members, groceries etc.) and to decide how they want to spend their newly freed up time.
The enterprise AI transformation journey is far more complex but the three phases remain the same. In previous articles we introduced the organizational work that needs to be done in the Hands Ready phase in terms of 7Rs: Reskill, Redeploy, Redesign, Reclaim, Recalibrate, Restructure and Remandate. These describe how the human organization must adapt as more operational control begins to move from people to AI – this is about relational transformation. But there is another side to the same squiggly line, a technological workstream running in parallel to the 7Rs. This is the work of helping agents become progressively more capable, more reliable, more integrated, more governable and more autonomous over time.
Production is just the start
In most enterprise technology programs, putting software into production is treated as the culmination of the roadmap. There may be adoption work after go-live. There may be training, support, bug fixes, enhancements and upgrades. But the basic assumption is that the system has been delivered and the primary goal of the program has been met.
Putting an agent into production, however, only marks the beginning of the Hands Ready phase. It means that AI is no longer merely assisting from the sidelines. It has been given a real operational role and is doing work that previously belonged to a human. But the fact that an agent is in production does not mean it is ready for Hands Off. It means the organization can now begin the work of progressively but intentionally releasing more control. In coordination with the iterative nature of the 7Rs, the technology work to be done is not a single implementation. It is a marathon of sprints, but where all the sprints are essentially the same, with each one asking whether the agent can now be trusted to operate inside a wider envelope than before. Can it handle more volume? More complex cases? More customer types? More systems? More steps in the workflow? More decisions? More exceptions? More collaboration with other agents? Less human review?
This is the main purpose of the technology part of the squiggly line, the iterative, cumulative expansion of the agent’s operating envelope. The operating envelope defines the conditions under which an agent is allowed to act. It includes the work the agent can perform, the cases it can handle, the systems it can access, the decisions it can make, the thresholds it must observe, the risks it can assume, the people or agents it can collaborate with, and the level of human oversight required.
At the beginning of Hands Ready, the operating envelope will usually be narrow. When first moved into production, an agent may be allowed to handle 20% of simple customer-service cases, or only internal requests, or only low-risk transactions, or only draft recommendations that a human must approve.
The next sprint might expand that envelope to 40% of simple cases. A later sprint might include moderately complex cases. Another might allow the agent to take action without approval below a certain financial threshold. Another might connect the agent to a pricing system, a billing system or another agent. Another might reduce human review from every output to exceptions only.
The normal expectation would be that the envelope expands with each successive sprint but there may be circumstances in which it needs to hold its current level or even to contract. The agent may be ready for more volume but not more complexity. It may handle more workflow steps but still require human approval for customer commitments. It may perform well alone but not yet collaborate safely with other agents. It may be technically correct but still weak in situations requiring trust, empathy or relationship judgment.
The srprint patten
To manage this expansion of the operating envelope, the team needs to ask itself a number of questions at the end of each sprint before it can progress onto the next:
The first question is simple: is the agent performing as expected at its current level? Has it delivered the value anticipated? Has quality held? Are errors, exceptions and escalations within acceptable limits? Are humans intervening more often than expected, or less? Are customers, employees or other users accepting the agent’s work? Has the agent stayed within its agreed boundaries? If the answer is no, the next move may not be expansion. It may be to hold the envelope steady, contract it, improve the context, redesign the workflow, adjust the governance model, or clarify the task contract.
If the answer is yes, the second question follows naturally: what additional part of the operating envelope should now be expanded? The expansion should be specific. It might mean moving from 20% to 40% of simple cases, from simple cases to moderately complex ones, from one workflow step to several, from recommendation to action with approval, from action with approval to action without approval, from one system to several, or from one agent acting alone to several agents collaborating. “Make the agent more autonomous” is not a sprint definition. “Allow the agent to resolve 40% of Tier 1 cases without human approval, excluding refunds above $500 and legally sensitive complaints” is.
The third question is: what additional value should this expansion create? The point of expanding the envelope is not simply that the agent does more work. It is that the agent creates more value by doing more work. That value may be cost reduction, speed, throughput, quality, revenue, conversion, customer experience, employee capacity, reclaimed demand, reduced risk or better decisions. Each sprint should therefore define not only what the agent will be allowed to do next, but why that greater freedom matters.
The fourth question is: does the agent have what it needs to succeed at the next level? This is where many agentic programs will either accelerate or stall. The agent may need better data, clearer policies, more reliable documents, richer customer history, stronger memory, access to additional systems, or a more precise understanding of business rules. It may also need the workflow around it to change. A process designed for human execution may not support agentic execution. Old handoffs, approvals, queues and review points may need to be removed, moved or redesigned. The agent cannot be expected to operate in a wider envelope if the business has not made that envelope legible and executable.
The fifth question is: what level of human governance is required now? In early sprints, humans may need to approve every action. Later, they may review selected actions, handle exceptions, audit samples or monitor patterns. Eventually, their role should move toward governing objectives, boundaries, risks and outcomes rather than checking individual outputs. This is one of the central transitions of Hands Ready. The goal is not to remove humans from governance, but to move humans to the right level of governance as the agent earns more operational control.
The sixth question is: what could go wrong, and how do we reverse it? Every expansion should define its risk thresholds, detection signals, escalation triggers, stop conditions and rollback plan. Some actions are reversible. Others are not. Some errors are tolerable. Others damage trust, create financial exposure or trigger regulatory risk. The ability to contract the operating envelope is what makes it possible to expand the envelope with confidence. Without rollback every expansion feels like a leap of faith. With rollback it becomes, almost paradoxically, a controlled release of control.
Together, these questions form the rhythm of the Hands Ready marathon:
Review the current envelope.
Decide whether expansion has been earned.
Define the next envelope.
Clarify the value.
Check the agent’s context and workflow.
Set the governance level.
Define the risk and rollback conditions.
Then run the sprint, gather evidence and begin again.
Each “squiggle” or wave in the line, each sprint, releases a little more control from human to agent. Each wave teaches the enterprise what the agent can do, what the workflow must become, where human judgment still matters, and what governance must now look like. Hands Off is reached when the agent has earned a wide enough operating envelope to complete a defined task, workflow or operating loop without routine human intervention. Not everywhere or all at once but deliberately, proof point by proof point, sprint by sprint.
From oversight to mission control
Across these sprints, the human role changes. At first, humans supervise the agent closely. They approve, correct, override and complete the work. They are still operationally close to the task. Over time, if the agent earns more autonomy, human attention should move upward. Humans handle exceptions. Then they review patterns. Then they govern boundaries. Then they set missions, objectives, values and constraints. They stop checking every turn of the wheel and start determining where the journey should go.
The Hands Off phase of autonomous transformation is the shift for humans from operational control to the higher level of mission control, from Hands Ready to Hands Off. Despite the fact that this is the goal of the AI transformation journey to the autonomous enterprise, we anticipate that many organizations will resist or put off the final push to Hands Off. Leaders may continue to intervene because intervention feels responsible. Managers may continue to require approvals because approvals preserve authority. Employees may continue to check the agent because checking gives them a sense of usefulness and safety.
This is why the technological side of Hands Ready cannot be separated from the organizational side. The agent’s operating envelope expands only as the organization’s confidence, governance, structure and human roles adapt around it.
The squiggly line re-visited
There are two complementary sides to the squiggly line of Autonomous transformation. On one side, the organization is iterating: re-skilling people, re-deploying talent, re-designing work, re-claiming value, re-calibrating metrics, re-structuring the enterprise, and re-mandating leaders.
On the other side, the technology is iterating: expanding the agent’s operating envelope, improving context, refining governance, reducing oversight, increasing authority, connecting agents and learning from exceptions.
These are two sides of the same transformation. The organization cannot reach Hands Off simply by deploying better agents. Nor can it reach Hands Off simply by changing roles and structures. It gets there when agents progressively earn more operational control and humans progressively learn how to release it. It is a marathon of sprints. Each sprint begins with the same question: what more can the agent now be trusted to do? Each sprint ends with another: what has to change before we release more control?
Eventually, for a bounded task, workflow or operating loop, the answer becomes clear. The agent can complete the work. The system is visible. The risks are governed. Exceptions are handled. Value is being created. Humans are no longer needed for routine intervention.
That is when the organization can move from Hands Ready to Hands Off. The squiggly line is the journey. The 7Rs and the operating envelope is how the journey is completed.