When AI is not enough

Column

Artificial Intelligence

July 29, 2026

When AI is not enough

Companies rehire the knowledge they tried to replace

The accelerated adoption of artificial intelligence in companies was accompanied by a simple, seductive, and dangerous promise: to replace people, reduce costs, and maintain, or even expand, the productive capacity of organizations.

This promise guided business speeches, investment decisions, internal reorganizations, and staff cuts across different sectors. In many cases, AI was presented not as a support tool, but as a direct alternative to human work, especially in areas considered repetitive, administrative, technical, or customer service.

However, operational reality begins to impose an important correction: automating tasks is not equivalent to replacing knowledge, experience, judgment, and professional responsibility.

This point is clearly illustrated in the recent Ford case. In June 2026, the company began to be cited as an example of a partial reversal of over-reliance on automated systems and AI tools. According to reports based on information from the company itself and accounts initially released by Bloomberg, Ford hired, promoted, or brought back about 300 to 350 experienced engineers, including former employees and professionals from suppliers, after realizing that automated quality systems and AI tools were not delivering, on their own, the expected level of reliability.

These professionals began to act in project reviews, early identification of failures, guidance of younger teams, and improvement of automated systems themselves (Ha, 2026).

The case is relevant because it does not indicate abandonment of artificial intelligence. On the contrary, it shows something more important: AI started to work better when it was placed back under the supervision of experienced professionals. Ford did not conclude that algorithms, computer vision, and automated tests were useless. The conclusion was different: automated tools depend on data, criteria, validation, context, and accumulated knowledge.

When the company reorganized its processes, hired veteran engineers, and strengthened the integration between engineering, manufacturing, supply chain, and quality, the results began to appear.

In 2026, Ford achieved leadership among mass-market brands in J.D. Power’s initial quality study, after years of underperformance. The company itself attributed this improvement to a combination of internal collaboration, technical review, hiring experienced engineers, and the use of automated support systems (Ford, 2026; J.D. Power, 2026).

Technical Rehire

The first important aspect of the Ford case is the recovery of the value of technical experience. For years, part of the business discourse on AI has treated professional knowledge as something easily capturable by data, documents, requirements, and statistical models. This view is limited. In complex industrial sectors, such as the automotive industry, the quality of a product does not depend solely on verifying whether a part meets an isolated specification.

It depends on the interaction between design, material, assembly, software, suppliers, actual use, maintenance, failure history, and accumulated decisions over successive development cycles.

Therefore, the rehiring of experienced engineers should not be read solely as a human resources move. It represents an epistemological correction. The company recognized that part of the knowledge necessary to produce quality was neither fully formalized in databases nor could it be replaced by the ingestion of requirements into AI systems.

Charles Poon, vice-president of vehicle hardware engineering at Ford, stated that AI is an important tool, but it depends on the quality of the information used in its training. He also acknowledged that the company had not paid enough attention to the experience of its most knowledgeable engineers, who had gone through many product cycles (Ha, 2026).

This recognition is central, because it exposes a common flaw in automation strategies: the confusion between information and competence. A documentary base can contain requirements, designs, standards, and failure records. This does not mean that it contains technical judgment.

The experienced professional does not just consult information. They interpret incomplete signals, recognize weak patterns, anticipate indirect effects, identify incompatibilities between areas, and perceive risks that have not yet appeared as measurable defects. AI can support this process, but it does not automatically replace the ability to formulate the correct problem.

In this sense, Ford’s experience is a practical response to the discourse that simply deploying AI is enough to replace qualified professionals. The company’s own quality outcome suggests the opposite.

The improvement occurred when the technical system was reorganized around collaboration, internal audit, engineering review, supplier integration, and the use of automated tools to support human work. The company did not improve because it chose between people and AI. It improved because it stopped treating AI as a sufficient substitute and began using it within a more mature organizational architecture.

Service and perceived quality

The same type of correction appears in another sector: customer service. Klarna, a Swedish financial services company, has become one of the best-known examples of an aggressive bet on AI for staff reduction.

In 2024, the company announced that its AI assistant was performing work equivalent to that of 700 customer service agents. The narrative was compelling, as it linked chatbots, cost reduction, and productivity gains. However, in 2025, the company hired people again for customer service roles and acknowledged the need to rebalance its strategy.

According to Reuters, CEO Sebastian Siemiatkowski stated that the company had “gone too far” in its emphasis on cost-cutting and was trying to correct course. The company also reopened vacancies after reducing its staff from 5,000 to 3,800 employees the previous year (Mukherjee; Wang, 2025).

The cases of Klarna and Ford are distinct, but they point to the same structural limit. In customer service, speed is not the only criterion of value. An automated response can reduce the average resolution time, standardize procedures, and decrease the cost per contact. However, the customer experience involves trust, empathy, negotiation, exceptions, ambiguity, frustration, and a sense of care. When a customer seeks support in a simple situation, an automated system may be sufficient.

When the problem involves conflict, financial loss, recurring error, urgency or contextual interpretation, the absence of human support can lead to a deterioration of perceived quality.

In May 2025, Customer Experience Dive reported that Klarna was rehiring people for customer service and that the company had come to advocate for the existence of a human option. The company’s own communication indicated a shift in emphasis: AI would offer speed, while human talent would offer empathy.

The formulation is relevant because it abandons the thesis of total substitution and adopts a logic of complementarity. Automating what is routine can be efficient. Eliminating the possibility of human intervention in complex situations can be a strategic error (Doerer, 2025).

This movement dismantles another frequent simplification: that quality can be reduced to short-term operational indicators. A chatbot can respond faster; a system can close more calls; a dashboard can show cost reduction, but none of this guarantees that the customer had their problem satisfactorily resolved, that the relationship with the brand was preserved, or that the system did not merely shift the cost to future complaints, loss of trust, cancellations, and reputational damage. AI can optimize local metrics while simultaneously worsening the overall outcome.

The error of direct substitution

On the other hand, rehiring professionals after AI-related layoffs should not be interpreted as a general failure of the technology. That would be another oversimplification. The problem is more specific and more serious: several companies seem to have adopted AI as a justification for layoff decisions before proving, at scale, that the tools would be capable of sustaining quality, continuity, and operational responsibility. In other words, AI was used as a narrative of efficiency before being validated as production infrastructure.

This difference is decisive. An AI demonstration can impress in a controlled environment. A pilot can generate punctual savings. A chatbot can answer frequently asked questions. A code assistant can produce useful snippets. However, companies do not operate solely on demonstrations.

They operate with exceptions, real clients, legacy systems, contracts, integration between areas, intermittent failures, incomplete data, regulatory decisions, reputational risk, and legal liability. The human work that seemed dispensable was often precisely the mechanism that absorbed these irregularities.

Gartner estimated in February 2026 that by 2027 half of the companies that attributed staff reductions in customer service to AI would rehire professionals for similar roles, albeit possibly with new titles.

The consultancy also highlighted that most recent workforce reductions stemmed not only from automation but from broader economic conditions, and that organizations would have to reinvest in human talent to sustain service quality and growth.

The analysis is important because it avoids both alarmism and propaganda. AI affects work, but does not automatically eliminate the need for expertise, empathy, and human judgment (Gartner, 2026).

Forrester reached a similar conclusion in January 2026. The company projected that automation and AI would have a real impact, but less than what the broad replacement discourse suggests. According to the consultancy, only 6% of jobs in the United States would be automated by 2030, while 20% of positions would be augmented by AI in the same period.

The report also indicated that more than half of the layoffs attributed to AI tend to be quietly reversed, as companies realize the operational difficulties of prematurely replacing human talent. Forrester itself warned of the risk of “AI washing”, that is, attributing financially motivated cuts to automation that is not yet mature enough to assume those functions (Forrester, 2026).

This point must be emphasized. Not every layoff announced in the name of AI means actual replacement by AI. In many cases, the technology serves as a public justification for restructurings involving falling demand, margin pressure, previous over-hiring, financial costs, or strategic repositioning. This does not make the problem smaller. On the contrary, it makes the situation more serious, because AI is used as a generic argument for decisions that do not always correspond to the actual technical capability of the implemented tools.

Organizational cost

The hasty replacement of people by AI also produces costs that do not immediately appear in the budget. The first is the loss of institutional memory. When a company lays off experienced professionals, it doesn’t just lose work hours. It loses knowledge about previous decisions, recurring problems, suppliers, clients, technical limitations, dangerous shortcuts, exceptions, and mistakes already made. This knowledge is rarely fully documented. Even when there is documentation, its usefulness depends on people capable of interpreting it.

The second cost is the degradation of supervision capacity. An organization that excessively reduces its technical staff may lose the competence needed to evaluate the AI it has come to use. This paradox is central. The more the company relies on automated tools to produce, review, serve, diagnose, or decide, the greater its internal capacity to validate results must be. If the organization cuts precisely the professionals capable of verifying errors, interpreting outputs, and correcting deviations, it increases its dependence on systems it does not fully understand.

The third cost is rework. AI can produce answers, texts, code, diagnoses, reports, and recommendations at high speed. However, production speed is not synonymous with delivered quality. In many contexts, initial savings are consumed by human review, error correction, discarding inadequate responses, reopening tickets, integration failures, and rebuilding trust. The company believes it has reduced costs because it decreased payroll, but part of the work reappears in a fragmented, distributed, and less visible way.

The fourth cost is rehiring. Rehiring after layoffs is not simply returning to the previous state. There are costs of search, integration, training, loss of continuity, drop in morale, and possible salary increases to attract professionals that the company itself had dismissed or devalued.

Furthermore, previous professionals are not always available, willing to return, or interested in rebuilding processes that were interrupted by hasty decisions. Correcting an automation error can cost more than the savings that justified it.

Market data reinforce that the phenomenon is not limited to isolated cases. In June 2026, Business Insider, based on a report by Challenger, Gray & Christmas, reported that AI had become the most cited reason by North American companies when announcing staff cuts.

In May 2026, AI was associated with 40% of the 97,006 layoffs announced by U.S. employers. Year-to-date, 87,714 layoffs had already been attributed to AI, up from 54,836 recorded in all of 2025. The report itself, however, highlighted that the causal relationship is contested and that some companies may be using AI as an explanation for broader adjustment decisions (Griffiths, 2026).

This dataset indicates a market in correction. On one hand, companies continue to announce cuts under the argument of AI. On the other hand, cases like Ford, Klarna, and projections from Gartner and Forrester indicate that a portion of these decisions will have to be revised. The contradiction reveals that the problem is not just in the technology, but in management. There are companies confusing tool with strategy, automation with competence, and staff reduction with productivity increase.

Productivity and actual value

The difficulty of converting AI into measurable economic value also appears in studies on business adoption. The report The GenAI Divide: State of AI in Business 2025, associated with MIT NANDA, pointed out that, despite estimated investments between US$ 30 and 40 billion in generative AI, about 95% of the organizations analyzed were not obtaining measurable returns.

According to the analysis released by Fortune, the main problem was not just the quality of the models, but a gap in learning and integration: generic tools work well for individuals, but tend to fail when they need to adapt to corporate workflows, internal processes, legacy systems, and organizational responsibilities (Estrada, 2025).

This diagnosis is directly related to the theme of rehiring. When AI is implemented without real integration into processes, it does not replace work. It creates an additional layer of apparent production. Companies start to generate more documents, more answers, more code, more reports, and more automated interactions, but they do not necessarily deliver more value. The increase in symbolic production can hide a drop in decision-making quality. Producing more does not mean solving better.

Real productivity depends on the fit between technology, process, and competence. An AI applied to a well-defined task, with adequate data, human supervision, and a clear evaluation criterion, can generate relevant gains. The same AI, applied to a poorly understood process, with bad data, vague goals, and absence of specialists, can only produce noise at scale. The error of many companies was assuming that the presence of the tool was enough. It is not enough. Tools enhance existing capabilities. When organizational capacity is fragile, they also amplify fragilities.

This distinction is decisive to avoid a mistaken reading of the critique. The issue is not to defend a return to pre-digital work, nor to deny that AI can generate productivity. The issue is to recognize that productivity is not born from the automatic replacement of people, but from the intelligent reorganization of work.

AI can reduce repetitive tasks, accelerate data analysis, recognize patterns, automate screenings, support diagnoses, and expand the capacity of qualified professionals. However, when used to eliminate knowledge before understanding the process, it can make the company faster at making mistakes.

AI Labor Governance

The main lesson from this cycle is that AI adoption must be treated as a governance decision, not an administrative fad. Before cutting people, the company must demonstrate that the technology supports equivalent or superior quality, that the risks have been mapped, that there is competent supervision, that indirect costs have been estimated, and that there is a plan for exceptions. The criterion should not be whether AI can perform a task in a demonstration, but whether it can operate reliably within the organization’s actual system.

This requires a change of question. Instead of asking “how many people does AI replace?”, the company should ask “which tasks can be automated without loss of quality, which require supervision, and which depend on specialized human judgment?”. The first question induces cuts. The second induces organizational design. The first caters to short-term enthusiasm. The second protects productive capacity in the medium term.

It is also necessary to distinguish three categories of work. There are repetitive and standardized tasks, in which automation can be dominant. There are semi-structured tasks, in which AI can prepare, organize, and suggest, but the decision must remain supervised. There are complex, ambiguous, critical, or relational tasks, in which direct substitution tends to produce high risk. Business failure occurs when these categories are treated as if they were equivalent.

In Ford’s case, the solution was not to abandon the technology, but to reintroduce specialists to guide, review, and improve the use of automation. In Klarna’s case, the fix was not to deny the chatbot, but to restore the possibility of human contact and shift the strategy from cost-cutting to quality and growth.

In the projections of Gartner and Forrester, the expected movement is not the disappearance of AI, but the partial reversal of hasty cuts and the creation of new roles associated with the supervision, integration, and responsible use of the technology.

Given this scenario, the necessary critique is simple: it is technically naive and managerially irresponsible to treat AI as a universal solution for productivity problems. AI does not automatically replace tacit knowledge, institutional memory, empathy, professional judgment, technical responsibility, and integration capacity. In some situations, it reduces cost. In others, it merely shifts cost. In others, it increases risk. In others, it reveals that the company did not understand its own process before trying to automate it.

In conclusion, the recent movement of rehiring and strategic correction shows that the most naive phase of business AI is beginning to encounter concrete limits. Companies that announced automation as a broad replacement for people are starting to face problems with quality, service, integration, supervision, and economic return. The technology will remain relevant, but the belief that it would solve everything is beginning to show itself as an expensive oversimplification.

The most likely future will not be composed of companies without professionals, operated by generic autonomous systems. It will be composed of organizations capable of combining AI with specialists, mature processes, reliable data, and governance. The company that understands this uses AI to amplify competence. The company that does not understand this lays off people, loses knowledge, discovers too late the value of what it cut, and then needs to rebuild, at a higher cost, the capability it believed it had automated.

To access the references of this text click here

Who wrote this column

Maurício Acconcia Dias

Possui graduação em Ciência da Computação pela Universidade Federal de Lavras, mestrado e doutorado em Ciências da Computação e Matemática Computacional pela Universidade de São Paulo e MBA em Data Science Analytics pela USP/Esalq. Atua com desenvolvimento de hardware para sistemas inteligentes aplicados à robótica. É consultor em Data Science & Analytics e Desenvolvimento de sistemas embarcados e orientador do MBA USP/Esalq.

You may also like

November 16, 2026

Exploratory modeling of the effect of preemptive cytomegalovirus treatment in AIDS patients

Cytomegalovirus (CMV) is highly prevalent and causes severe disease in immunocompromised individuals, such as HIV patients. Given the scarcity of robust data on preemptive treatment for this infection, an exploratory model was developed. The objective was to identify potential subgroups that would benefit from this approach. A retrospective observational study was conducted, analyzing 176 medical records of patients admitted to a tertiary hospital between 2017 and 2022. After applying inclusion and exclusion criteria, 116 patients were divided into a preemptive treatment group (19) and a control group (97). Machine learning techniques, such as LinearDML and CausalForestDML, and variable analysis by Feature Importance and SHAP were used. Variable analysis indicated that high CMV viremia was associated with death, with an Odds Ratio of 1.85 (p = 0.004). The model estimated an increase in overall mortality associated with preemptive treatment (ATE of 48%, p = 0.0180), possibly reflecting the greater baseline severity of the treated patients. However, heterogeneity analysis identified that older patients showed a relatively more favorable estimated effect. Limitations included the small sample size and the retrospective and observational nature. Nevertheless, with an expanded evaluation group, the model could serve as a basis for building clinical decision support tools.

Keywords: Machine learning; Double machine learning; Causal inference; Mortality.

November 16, 2026

Regression models for personal hygiene and personal care product expenses in the Southeast region

Regression models were evaluated to correlate expenses with personal hygiene and care products of the population in the Southeast Region of Brazil. Data from the 2017-2018 Family Budgets Survey (POF), conducted by the Brazilian Institute of Geography and Statistics, were used. The models were estimated by ordinary least squares and verified for the assumptions of normality of error terms, multicollinearity, and heteroscedasticity. Linear regression, nonlinear regression with Box-Cox transformation, and linear regression with dummy variables were evaluated. Coefficients statistically different from zero were obtained for all three models, confirmed by the p-values of the F and t tests, and the adherence of residuals to normality was graphically assessed. However, heteroscedasticity problems were observed in all cases, indicating the omission of relevant variables from the model. The low R² values obtained, between 2.19% and 5.56%, suggested that the dependent variable was poorly explained by the models. Despite this, a growing trend in expenses for personal hygiene and care as a function of disposable income per capita was observed. For future studies, it is recommended to evaluate other explanatory variables to improve the model’s adjustment coefficient and eliminate heteroscedasticity.

Keywords: Family expenses; Ordinary least squares; Per capita income; Box-Cox transformation; Dummy variables.

November 16, 2026

Cognitive Accessibility and ASD: Design Guidelines for Augmentative and Alternative Communication Applications

Autism Spectrum Disorder (ASD) imposes specific challenges on the design of digital interfaces, affecting cognitive processing, sensory regulation, and functional communication. In this context, Augmentative and Alternative Communication (AAC) is an essential resource to support individuals with oral language limitations. This study aimed to analyze the usability and accessibility of AAC applications for the neurodivergent public. The research was characterized as descriptive, with a quantitative approach, and was based on the systematic observation of the interfaces of the Brazilian applications Livox, Matraquinha, and Expressia. A structured checklist, developed from the convergence of Human-Computer Interaction (HCI) guidelines, Applied Behavior Analysis (ABA), and the GAIA Guide, was used as a data collection instrument. The results revealed that Livox and Expressia showed greater technical compliance and learning support, while Matraquinha stood out for its semantic literalness, but with limitations in customization and architecture. A generalized critical gap was identified in the proactive detection of frustration, highlighting opportunities for strategic innovation for the digital business market and for the advancement of neuroinclusive design.

Keywords: Neuroinclusive Design; Digital Inclusion; Strategic Innovation; Assistive Technology; Usability.

November 16, 2026

Multidimensional Poverty and Bolsa Família Program: challenges and perspectives of inclusion through public policies

The concept of poverty has evolved over the 20th and 21st centuries, incorporating a more inclusive perspective that transcends monetary income. The Multidimensional Poverty Index (MPI) has emerged as a complementary tool to characterize poverty and guide programs like Bolsa Família, aiming to improve their effectiveness and targeting. The study aimed to create a national MPI using data from the Continuous National Household Sample Survey (PNADC) of 2024, seeking to identify the correlation between receiving the benefit and multidimensional poverty, as well as to detect underserved populations. The Alkire-Foster methodology was used to calculate the MPI, decomposing the results into groups of beneficiaries and non-beneficiaries of PBF. Subsequently, a Generalized Linear Model with a logit function was applied to analyze the correlations. The results revealed that 4.84% of the Brazilian population lived in multidimensional poverty, with an incidence three times higher among Bolsa Família beneficiaries. Two profiles of multidimensional poverty and two potential blind spots in PBF coverage were identified, including 9.0% of non-beneficiary multidimensional poor in extreme monetary poverty and 25.2% in low-income vulnerability. Being a PBF beneficiary was associated with a two-fold higher chance of being classified as multidimensionally poor. It was concluded that the MPI is a valuable tool for designing public policies, increasing targeting capacity and demanding actions to address structural problems that perpetuate poverty.

Keywords: Alkire-Foster; MPI; Public policies.

November 16, 2026

The influence of People Management on the Health and Well-being of collaborators

In a dynamic organizational scenario, understanding the influence of People Management practices on employee health and well-being has become essential. The study aimed to analyze how People Management practices influence employees’ perception of health, well-being, and quality of life in various organizations. The methodology adopted was descriptive, with a quantitative approach, and employed the survey technique. Data were collected through a structured questionnaire with 15 sociodemographic questions and Likert scale items, disseminated for 40 days on social media. The sample consisted of employees from different companies, selected for convenience, with voluntary and anonymous participation. The results revealed that people management practices impacted motivation, engagement, and the perception of quality of work life, but were not consistently perceived by all employees. It was observed that organizational actions focused on health and well-being were recognized in a moderate and non-equivalent manner, with the presence of neutral and dissenting perceptions. It was concluded that, although People Management practices have strategic potential, their effectiveness depends on greater structuring, continuity, and communication. The study contributed to highlighting gaps between the adopted practices and employee perception, providing subsidies for the improvement of organizational strategies and for the production of academic knowledge in the area.

Keywords: Organizational well-being; Organizational climate; Engagement; Motivation; Quality of work life.

November 16, 2026

Optimization of Delivery Logistics Network through Mathematical Modeling and Clustering

Efficiency in distribution logistics constitutes an essential competitive advantage for the reduction of operational costs and the financial sustainability of contemporary organizations. The general objective of this work was to optimize the delivery network of a company, aiming at cost reduction through the decrease of mileage traveled. The methodology employed was based on the principles of Operational Research and was structured in three sequential phases: geographical allocation of clients to depots respecting Federal Unit restrictions, clustering of stores using the K-Means algorithm, and routing of the heterogeneous fleet using the mathematical formulation of the Vehicle Routing Problem. The main results obtained demonstrated a reduction in the total distance traveled, which represented an absolute saving of 5,116 km and a percentage gain of 3.24%. The model validated itself as a robust tool, capable of optimizing vehicle occupancy and eliminating subjectivity in logistical planning. It was concluded that the use of advanced mathematical modeling allowed operational restrictions to be converted into a measurable competitive advantage. The solution validated the viability of data-based decision support systems for the automation and optimization of large-scale logistical flows.

Keywords: Clustering; Logistics optimization; Operational Research; Cost Reduction; Vehicle routing.