The 99% figure should worry boards for a reason most executives are missing
A recent discussion around Mercer Global Talent Trends data reported an extraordinary signal: 99% of surveyed CEOs expect AI initiatives to lead to layoffs in the near term. That number is dramatic, but it is not the most important part.
The more revealing figure is that only 32% believe human and machine capabilities can be optimally combined in the workforce.
That gap tells us something uncomfortable. Many organizations are not yet treating AI as a serious operating model transformation. They are treating it as a headcount narrative.
AI can absolutely reduce labor intensity. But if the main strategic idea is simply to replace people, the organization has probably not understood AI well enough to benefit from it.
The question boards should ask is not, How many roles can we cut? The better question is, Which decisions, workflows, controls, and capabilities can we redesign so one capable person can supervise and improve hundreds of AI-supported processes?
That is a very different management conversation.
AI is not a technical deployment. It is a redesign of judgment
The common mistake is to place AI in the same mental category as software automation. Traditional automation works best when the process is deterministic: if X happens, do Y. AI is different. Its real value appears when organizations can execute non-deterministic processes, where judgment, language, interpretation, prioritization, and context matter.
That means AI can support work that used to require human discretion. Contract review, customer response triage, claims assessment, code generation, procurement analysis, regulatory summarization, internal knowledge search, financial commentary, sales preparation, quality assurance, and operational exception handling are all examples.
But this does not mean humans disappear. It means the human role changes.
A weak implementation puts a human approval step after every AI action and calls that governance. That creates expensive friction and delivers very little operational leverage.
A strong implementation asks a sharper question: how do we move the human from being the operator of one process to being the supervisor, trainer, and exception manager of many processes?
This is where most AI strategies succeed or fail.
Layoffs may happen, but they are a poor measure of AI maturity
Cost reduction is a legitimate business objective. Some roles will change. Some will shrink. Some will disappear. Pretending otherwise is not serious.
Still, layoffs are not proof that an AI program is working. They may simply prove that management found a new justification for an old cost-cutting plan.
Real AI maturity should be measured through operational and financial indicators such as:
- Cycle time reduction in core processes
- Lower error rates in high-volume judgment tasks
- Higher throughput per expert employee
- Reduced dependency on manual coordination
- Faster onboarding of new employees into complex workflows
- Better quality of decisions due to richer context
- Stronger auditability of human and machine actions
- Revenue protection through faster response and detection
If the only measurable outcome is fewer employees, the company may be harvesting short-term savings while damaging long-term capability.
The hidden risk: destroying the talent pipeline
The most exposed group is often early-career talent. Many entry-level tasks are exactly the tasks AI can now perform reasonably well: summarizing, preparing drafts, checking documents, basic analysis, simple coding, support triage, research assistance, and operational follow-up.
The managerial temptation is obvious. Why hire juniors if AI can do much of the junior work?
The problem is that junior work has never only been about output. It is also how organizations create future experts. People learn the business by doing the basic work first. They develop pattern recognition, professional discipline, domain intuition, and judgment through exposure.
If companies remove the first rung of the ladder, they should not be surprised when five years later they cannot find people ready for the third rung.
This is especially relevant in markets such as Israel, where high-performing technology companies depend heavily on dense professional networks, strong engineering culture, and fast learning loops. A short-term reduction in junior hiring can quietly weaken the future management bench, product intuition, and technical depth of the organization.
AIRD is not just an employee psychology problem
Some researchers have started using the term AIRD, or AI Replacement Dysfunction, to describe the anxiety and performance decline caused by fear of AI-driven replacement.
Executives may be tempted to dismiss this as emotional resistance. That would be a mistake.
When employees believe AI is mainly a weapon against them, three things happen:
- They hide knowledge instead of sharing it
- They resist adoption even when tools are useful
- They stop investing in long-term organizational learning
This creates a paradox. The company wants AI to learn from its best processes, but the people who understand those processes no longer trust the transformation.
Successful AI adoption requires operational trust. Not softness. Not vague inspiration. Trust built through clear governance, honest communication, incentives, and visible investment in employee capability.
The two-track AI strategy every serious company needs
Organizations should not choose between AI literacy and AI agents. They need both.
The first track is broad AI literacy. Employees must learn how to communicate effectively with models, validate outputs, structure requests, manage context, and understand the limits of AI-generated work. This is now a core professional skill, similar to spreadsheet literacy in previous decades.
The second track is agent development. Companies need internal capability to design, deploy, monitor, and improve AI agents that execute real processes across systems.
These tracks are different.
AI tools often require employees to change habits. They need to remember to use the tool, understand when it helps, and adapt their personal workflow. That can be harder than it looks.
AI agents, by contrast, can often be embedded into the operating environment with less behavioral change from employees. The technical architecture may be more complex, but the adoption path can be simpler because the agent works inside an existing process.
This is why the future enterprise needs a platform for creating and managing agents, not just a collection of disconnected AI subscriptions.
IT departments will become HR departments for AI agents
A serious enterprise AI program needs more than experimentation. It needs an operating layer for agents.
That layer should answer practical questions:
- Who owns each agent?
- What data can it access?
- What actions can it perform?
- How is performance measured?
- When does it escalate to a human?
- How are failures reviewed?
- How are prompts, tools, permissions, and model versions governed?
- How are agents retired when they are no longer useful?
This is why information systems teams will increasingly act like human resources departments for AI agents. They will onboard, permission, evaluate, supervise, and offboard digital workers.
The companies that build this capability early will move faster than companies that treat every AI use case as a one-off pilot.
Tool selection matters, but it is not the strategy
The current enterprise AI stack is becoming more interesting. Claude is one of the strongest systems for broad organizational adoption, especially where reasoning, writing, analysis, and coding support matter. Claude Code and collaborative Claude-based workflows are among the more practical AI capabilities available today, although security and data governance must be handled carefully.
Microsoft Copilot is a solid infrastructure choice for organizations already deep in the Microsoft ecosystem. It has sometimes moved slower than smaller AI-native competitors, but recent improvements are meaningful, and Copilot Studio can be useful for Microsoft-centered agent scenarios.
At the same time, tools such as n8n are entering environments that once would have considered them too unconventional for large enterprises. This is an important signal. The enterprise automation layer is becoming more open, composable, and agent-friendly.
Still, no tool compensates for weak process thinking. A company can buy excellent AI products and still fail if it lacks domain expertise, governance, and a realistic model of how work actually happens.
Beware the self-appointed AI expert
AI is now attracting a large number of opportunistic advisors. Some have impressive social media presence but limited business experience, limited academic depth, and little understanding of how organizations actually operate.
Large enterprises can usually filter this noise. Small and mid-sized businesses are more vulnerable. They may adopt fragile automations, expose sensitive data, overestimate model reliability, or redesign processes based on shallow advice.
This matters because AI is a multidisciplinary field. It requires technical understanding, but not only technical understanding. It also requires management experience, process design, domain knowledge, risk awareness, and in many cases academic seriousness.
The best AI work often comes from people who can connect research, business operations, and implementation. Not from people who merely know the newest tool.
A better executive playbook for AI workforce transformation
If CEOs expect layoffs, they should also be expected to present a more disciplined AI workforce plan. That plan should include six elements:
- Map judgment-heavy processes, not only repetitive tasks.
- Identify where AI can increase expert leverage rather than merely remove junior work.
- Define human-in-the-loop patterns that scale, with humans supervising portfolios of AI activity rather than approving every micro-action.
- Build internal agent development capability, including governance, monitoring, security, and ownership.
- Invest in AI literacy across the workforce, especially model communication, validation, and process redesign.
- Protect the talent pipeline by redesigning entry-level roles instead of eliminating them blindly.
This is the difference between AI as an operating advantage and AI as a layoff slogan.
The real board-level question
The 99% statistic is not just a labor market warning. It is a governance warning.
Boards should challenge management teams that present AI transformation only through savings and headcount reduction. Savings matter, but they are not enough. AI should also improve resilience, quality, speed, knowledge retention, customer experience, and decision-making.
If an organization cuts people before it understands the processes those people operate, it may lose precisely the knowledge required to automate intelligently.
The winners will not be the companies that cut fastest. They will be the companies that redesign work with enough depth to make both humans and machines more productive.
AI will change the workforce. That part is no longer controversial. The open question is whether leaders will use it as a blunt instrument or as a serious management discipline.
