AI Did Not Fail — Poor Workforce Strategy Did
The recent pattern of organisations laying off staff because of AI, then rehiring people months later, should not be read as proof that AI has failed. It is better understood as proof that rushed workforce strategy has consequences.
Introduction
A Careerminds survey of 600 HR professionals who had made layoffs in the previous 12 months found that many organisations were already reversing course. Among companies that had conducted AI-led layoffs, 32.7% had rehired between 25% and 50% of the roles they initially cut, while 35.6% had rehired more than half of those roles. When rehiring occurred, it often happened quickly: 52.1% of HR leaders said their organisation rehired for previously eliminated roles within six months.
This is not an argument against AI. It is an argument against simplistic thinking.
AI can automate tasks, accelerate analysis, support software development, improve customer operations, and reduce administrative friction. But a task is not the same as a role, and a role is not the same as the human judgement, institutional memory, relationships, and accountability that sit behind it.
The lesson is clear: the future of work is not AI instead of people. It is AI-literate people working with well-designed AI systems.
The Problem: AI Capability Does Not Automatically Mean Workforce Replaceability
The major misconception is that AI capability automatically equals workforce replaceability.
Many organisations looked at generative AI, automation platforms, and agentic tools and saw an opportunity to reduce cost. In some cases, that may have been reasonable. Repetitive workflows, routine support tasks, basic reporting, content drafting, and internal knowledge retrieval can often be improved with AI.
The problem begins when leaders move from “this task can be automated” to “this person is no longer needed”.
Visible work
Most jobs contain visible work that is easier to document: responding to tickets, writing reports, testing code, processing requests, creating content, and producing analysis.
Invisible work
Most jobs also contain invisible work that is harder to measure: knowing when a process is about to fail, understanding why a customer is frustrated, remembering how a product behaved three releases ago, spotting a weak assumption in a business case, or knowing which stakeholder needs to be consulted before a decision is made.
AI systems are not useless in these environments. They can be very helpful. But they need boundaries, context, validation, escalation pathways, and human accountability.
Klarna: A Useful Example
Klarna is a useful example of this shift. Reuters reported that the company’s CEO said Klarna may have gone too far in using AI for cost-cutting and was shifting focus toward improving services and products. The company had reduced staff and leaned on AI for customer queries, but later returned to hiring and reframed AI as more than a cost-cutting tool.
That is the strategic point. AI used only as a cost reduction instrument often produces a narrow view of value. AI used as a capability multiplier can produce a more durable business outcome.
Analysis: Why the AI Boomerang Effect Is Not Surprising
The so-called AI boomerang effect is not surprising when we look at how organisations actually work.
1. Complex work rarely follows a clean script
Customer service, engineering, HR, compliance, sales, operations, and technology delivery all involve exceptions. These exceptions are where experienced people matter most. A chatbot may answer common questions well, but it may struggle when the customer is emotional, the issue crosses multiple systems, or the policy is ambiguous.
2. Institutional memory is an operational asset
Institutional memory does not sit neatly in a database. It lives in the experience of people who know why decisions were made, which shortcuts are dangerous, which systems are fragile, and which customers, suppliers, or internal teams require careful handling.
Ford’s recent quality reset illustrates this point. Business Insider reported that Ford hired, promoted, or brought back about 350 experienced technical specialists to help fix vehicle-quality problems. Ford executives acknowledged that AI and automation were not enough on their own, with one executive noting that AI is only as good as the information used to train it.
3. AI creates new work even when it removes old work
Outputs must be checked. Errors must be corrected. Prompts must be refined. Workflows must be redesigned. Risks must be monitored. Data must be governed. People must decide when an AI-generated answer is good enough and when it needs escalation.
Careerminds found that 54.6% of surveyed companies felt AI-led layoffs were not worth it because the technology required more human oversight than expected. The same research found that 32.9% of HR professionals believed their organisation had lost critical skills and expertise through layoffs.
Cutting people before understanding the operating model can make the organisation less efficient, not more.
Practical Implications
For business leaders
AI transformation must start with work design, not headcount reduction.
Before removing roles, organisations should map the work. Which tasks are repetitive and low-risk? Which tasks require human judgement? Which tasks involve regulated decisions, customer trust, safety, security, ethics, or brand reputation? Which tasks can AI assist but not own?
For employees
AI literacy is becoming a core career skill. This does not mean every person must become a machine learning engineer. It means people need to understand how AI tools work, where they are useful, where they are unreliable, and how to use them safely in their own context.
For technical teams
AI implementation requires more than tool deployment. It requires evaluation, monitoring, data quality, access control, escalation design, auditability, and ongoing human review.
For HR and workforce planners
Reskilling should come before redundancy. Careerminds found that 55.1% of companies had not formally discussed or considered reskilling and redeployment before AI-related redundancies. Yet many later believed some roles could have been transitioned with the right support.
That is a missed opportunity. Organisations that remove people too quickly may later pay twice: once through severance, disruption, and lost knowledge; and again through rehiring, retraining, and rebuilding trust.
Strategic Advice
A more pragmatic approach is to treat AI adoption as a capability programme, not a cost-cutting campaign.
1. Create a human-in-the-loop operating model
Decide where AI can act independently, where it can recommend, where it must be reviewed, and where it should not be used at all.
2. Invest in AI literacy across the workforce
Leaders need enough understanding to make responsible decisions. Managers need to redesign workflows intelligently. Employees need to use AI confidently and critically. Technical teams need deeper expertise in integration, data governance, evaluation, and risk management.
3. Protect institutional knowledge
Before restructuring, organisations should identify critical expertise, document key decision logic, and involve experienced employees in training, validating, and improving AI-enabled workflows.
4. Measure value properly
Labour cost reduction is only one metric. Organisations should also measure quality, customer satisfaction, cycle time, employee experience, risk exposure, rework, compliance, and long-term capability.
5. Avoid extremes
The choice is not between “replace everyone with AI” and “avoid AI altogether”. Both positions are weak. The stronger position is disciplined optimism: use AI where it genuinely improves the work, and keep humans accountable where judgement, care, context, and responsibility matter.
IBM’s public description of AskHR reflects this more balanced direction. The company describes the model as hybrid automation, with digital labour supporting zero-touch HR tasks while preserving the human experience where needed.
That is the direction more organisations should take.
Conclusion
The AI layoff reversal story is not a warning to stop using AI. It is a warning to stop making shallow assumptions about work.
AI can be powerful, but it does not remove the need for judgement. It can automate tasks, but it does not automatically replace roles. It can increase productivity, but only when implemented with governance, training, redesign, and accountability.
The organisations that succeed will not be the ones that simply cut faster. They will be the ones that learn faster.
They will build AI literacy across their workforce. They will redesign roles rather than delete them reflexively. They will use human-in-the-loop systems where the risk requires it. They will understand that institutional memory is not a cost centre; it is part of the organisation’s operating intelligence.
AI did not fail. Poor workforce strategy did.
The way forward is not panic, resistance, or blind optimism. The way forward is a practical partnership between people and technology, where humans remain responsible for meaning, judgement, trust, and direction — and AI becomes a tool that helps them do better work.
Written by
Adil Bilal
Founder & Principal Consultant
Adil Bilal is an AI Consultant and Engineer specialising in Generative AI, Agentic AI systems, Large Language Models (LLMs), Vision-Language Models (VLMs), data analytics, and AI governance, helping organisations design and implement trusted, explainable, and business-aligned AI solutions.
