All writing
·Saahil Panikar

43% of Leaders Expect Significant Job Disruption in the Next 18 Months. They're Not the Ones Whose Jobs Are at Risk.

AI LeadershipAI GovernanceAgentic AIWorkforce Transformation

Forty-three percent of business leaders expect a lot to extreme job disruption in the next 12 to 18 months because of AI agents. That number comes from Deloitte, published this August, from a survey of 501 senior managers and C-suite leaders. The leaders predicting the disruption are the ones deploying the agents. They are not, for the most part, the ones whose jobs will change. That is a structural accountability gap, and most leadership teams haven’t designed around it.

The Deployment Is Already Ahead of the Readiness

Forty percent of respondents from organizations with more than $1 billion in revenue report scaling AI agents, up from 27% a year ago (McKinsey, August 2026). Agentic AI systems that make multi-step decisions, execute tasks autonomously, and act on behalf of humans across entire workflows are live in production environments today. In your industry, and quite possibly in your company.

Only 16% of organizations say their processes are ready for them. Five percent describe themselves as highly prepared.

Scale doesn’t close the gap. Even among organizations already running agents at scale, only 46% say their business processes are prepared. The rest have deployed autonomous systems into processes that weren’t designed for them, aren’t governed for them, and haven’t been tested against the scenarios those agents will actually encounter. They have deployed capability faster than their organizational structure can absorb it.

This is not a technology problem. The agents execute. The models work. The failure is structural, and it lives upstream, in the design decisions leaders made, or didn’t make, before the first agent went live in a workflow that touched people’s livelihoods.

“Unready” Is Not a Technical State. It’s an Organizational One.

When I work inside defense, aerospace, financial services, and energy organizations, the readiness gap is never the AI system itself. The use case is real. The integration is functional. What’s missing is almost always the same set of things.

No defined decision rights for what the agent can and can’t do autonomously. No explicit human-AI handoff protocols identifying where human judgment must re-enter the loop. No accountability structure for outcomes when the agent acts without a human in the approval chain. No validation gate between agent output and consequential action.

In a commercial software environment, those gaps produce inefficiency. You deploy something underperforming, you iterate, you correct. The cost is time and money, and it’s recoverable.

In a defense program, a financial services compliance function, or an aerospace supply chain, those same gaps produce a different category of problem. A missing human review step in a contract compliance workflow isn’t an efficiency loss. It’s a regulatory risk. An agent operating without defined decision rights in a process that feeds safety-critical program delivery isn’t a productivity gap. It’s a mission risk.

I have seen both. The cost of getting this wrong in regulated, mission-critical environments doesn’t fit on a two-week iteration cycle.

And in every industry, when agents operate in undefined processes, the collateral damage doesn’t hit the system log first. It hits the people in the roles adjacent to the automation: the ones who weren’t told exactly how their work was changing, whose job descriptions haven’t caught up to what the agent is now handling, and who find out what their role has become after the deployment is already in production.

The Leaders Making the Deployment Decisions Aren’t the Ones Living With the Outcomes

The leaders deciding to scale agents are generally not in the roles that will be most disrupted by them. They are making investment and deployment decisions with significant workforce implications from positions that are, at most, indirectly affected. That isn’t an indictment of individual leaders. It’s a description of an accountability structure designed for a different era.

What makes it a governance failure is the combination: deployment at speed, without the organizational design that would give the people most affected any meaningful visibility, voice, or protection in the process.

Three-quarters of the leaders in the Deloitte survey see more value in human-agent collaboration than in pure automation. That’s the right instinct. But only 45% have defined the human-agent operating model to achieve it, and half say their organizations aren’t investing enough in the workforce transformation that collaboration requires.

Most leaders understand, at least in the abstract, that partnership between humans and agents is the answer. Far fewer have built the structure that makes it function when the agent acts autonomously, something goes wrong, and someone needs to know who owns the outcome, who can override it, and what happens next. The gap between what leaders believe and what they have actually designed is where the disruption they are predicting originates.

The Readiness Work Produces Value That Outlasts the Agent

I worked with a large aerospace and defense organization deploying agents across several program management functions. The technical implementation was not the hard part.

Before a single agent went live in a workflow that touched human roles, we mapped every decision point the agent would encounter. We defined explicit human-AI handoffs: which outputs required human review before action, which could proceed autonomously within defined parameters, and under what conditions the agent would escalate rather than act. We assigned an accountable owner for each handoff. We built a validation gate, and we ran the design against a set of realistic failure scenarios before anything reached production.

What came out on the other side was not just a safer deployment. It was a clearer workflow than the one that existed before the agent arrived. Designing for the agent forced the organization to answer process questions it had been deferring for years. Delivery predictability on the affected programs improved by 62% in the two quarters that followed, a direct consequence of the workflow clarity that preceded deployment, not of the AI capability itself.

That pattern repeats. The organizations getting real value from agentic AI aren’t the ones who moved fastest. They’re the ones who defined the process before the agent entered it.

The Governance Model Has a Name Now

AI-Native SAFe, which Scaled Agile opened for early access in June and announced for general release at SAFe Summit in San Diego this month, is an enterprise agility framework built to address this specific gap at scale. It doesn’t replace the Lean and Agile investments organizations have made in the Scaled Agile Framework (SAFe) over the last decade. It builds on them, and it adds what the agentic era specifically demands.

The framework formalizes what I see the strongest organizations already doing: human-AI handoffs designed into the operating model rather than retrofitted after deployment, an AI Value Architect role that guides teams toward outcomes while managing cost, ethics, legal, and risk, and a validation-first orientation. As Andrew Sales, Scaled Agile’s chief methodologist, put it, the challenge is no longer whether organizations can build something in the time available: “it is keeping up with the need to validate whether what they are building is safe, secure, and valuable.”

That reframe matters in every industry. It matters most in the ones where the consequences of unvalidated agent action aren’t recoverable on the next sprint.

Three Decisions Have to Come Before the Next Deployment

Before your next agent deployment expands, three things need to be defined. They belong in a leadership conversation, not an implementation ticket.

Decision rights. What can the agent do autonomously, within what parameters, and who owns the outcome when it acts? This is an organizational commitment, not a configuration setting, and it has to come before deployment, not after the first incident.

Human-AI handoff design. At what points does human judgment re-enter the loop, and who is responsible for it? In mission-critical and highly regulated environments, this is the design constraint the deployment must be built around, not a question answered after the fact.

Workforce transition accountability. Who in your leadership structure owns the obligation to the people whose roles are changing? Not the training rollout. Not the change management communication plan. The actual accountability for what those individuals do next and whether the organization supported them in getting there.

The organizations that answer these questions before scaling don’t move slower. They move with more precision. And when something goes wrong, as something eventually will, they have the structure to identify it, contain it, and correct it before it becomes the kind of workforce story no communication plan can manage.

Forty-three percent of leaders expect significant job disruption in the next 18 months. That prediction tells you what most of them believe is coming. What it doesn’t tell you is whether any of them built the governance model that makes them accountable for how it lands on the people doing the work.

That is the decision still in front of you.

AI won’t transform your company. Leaders will, with the right operating model behind them.

For the data on how the enterprise AI bottleneck shifted from budget and skills to governance and decision rights, see The AI Bottleneck Has Moved. Have You?

Sources

Work With Atlas

If the stakes are high, let's talk.

Atlas Revolutions partners with a small number of Fortune 100 and government leaders at a time. If you're navigating a transformation where getting it wrong isn't an option, start with a conversation.

Start the Conversation