"Big Tech Walks Back AI Replacement Narrative, Bets on AI That Works Alongside Humans"

Meta AI Executive: "AI Is Not Meant to Replace Humans but Serve as an Economic Partner"
Real-World Industry Evaluations Confirm Strength in Repetitive Tasks, Limitations in Specialized Work
Rising Enterprise Costs Prompt Strategic Reassessment of AI Deployment

The role of generative artificial intelligence (AI) is increasingly being redefined from technology that replaces human workers to a workplace partner that collaborates with them. Real-world evaluations of AI deployed in industrial settings have demonstrated strong performance in repetitive tasks and information processing while continuing to expose limitations in complex, specialized work. As companies adopting AI recalibrate deployment strategies around cost efficiency, leading AI developers including Meta and Anthropic are likewise sharpening their focus on AI assistants and enterprise-oriented services.

The Next Frontier of AI: Delivering Tangible Economic Value

According to the South China Morning Post (SCMP) on June 30, Dawn Song, the executive recruited to lead Meta's newly established Superintelligence organization, outlined a pragmatic vision for AI's evolution during an interview conducted at the World Economic Forum (WEF) in Dalian, China, shortly before joining Meta. Song said, "The ultimate goal of AI is not to take away and replace human jobs," adding, "What we want is for these intelligent assistants to help people perform their work more effectively in real industrial settings, thereby creating tangible economic value for businesses."

Meta's AI development strategy has increasingly aligned with that vision. Rather than building systems designed to replace people, the company is concentrating on creating AI that works alongside humans while devoting substantial research resources to ensuring safety and reliability. Under this model, AI handles repetitive and standardized tasks such as drafting emails, conducting research, managing schedules, organizing documents, and writing code, allowing people to focus on judgment, decision-making, and creative work. Meta's decision to recruit Song, a globally recognized authority on AI safety, reflects the same strategic rationale. As AI assumes a larger share of real-world work, accuracy, reliability, and safety increasingly become core determinants of competitiveness.

The rationale behind Meta's human-centric AI strategy is also reflected in recent evaluations of AI agents. Earlier this month, the Responsible Distributed Intelligence (RDI) Center at the University of California, Berkeley, released the Agents' Last Exam (ALE), an evaluation framework designed to measure AI agents' ability to complete real-world work. The benchmark assesses whether AI can perform approximately 1,500 authentic tasks collected across 55 professional industries at a level comparable to humans. Test assignments included operating professional video editing software to complete editing projects and interpreting brain magnetic resonance imaging (MRI) scans, emphasizing practical job execution rather than measuring knowledge alone.

The results closely aligned with Meta's strategic direction. State-of-the-art AI models performed well in repetitive work and information processing but recorded significantly lower completion rates on lengthy, multi-step assignments and tasks requiring specialized expertise. Song noted, "The problems in ALE are extremely challenging and realistically designed, and even the world's most advanced AI models, widely regarded as the smartest available today, achieved very low passing rates."

Competition Centered on Workforce Efficiency

The redefinition of AI's role is unfolding alongside broader changes in the labor market. Over the past two to three years, companies have increasingly used generative AI as a cost-reduction tool. Automation of repetitive work, reductions in white-collar staffing, and streamlined management structures have advanced simultaneously. In a memo to employees last June, Amazon CEO Andy Jassy said the expansion of generative AI and AI agents could reduce the size of the company's corporate workforce over the coming years. At the time, Amazon had more than 1,000 AI services and applications either deployed or under development. Jassy told employees that "AI will change how work gets done" and urged every organization to strengthen its AI capabilities.

The wave of workforce reductions across Big Tech has emerged under similar pressures. Microsoft (MS) launched a restructuring initiative eliminating roughly 9,000 positions, or about 4% of its global workforce, as AI infrastructure expansion costs continued to rise. According to Reuters, Microsoft has simultaneously streamlined its organizational structure and reduced management layers to absorb mounting investments in cloud and AI infrastructure. Oracle also reduced its workforce by 21,000 employees during fiscal 2026 as it reshaped operations around AI and cloud infrastructure, while restructuring expenses climbed sharply to $1.8 billion.

The repeated discussions surrounding workforce reductions have largely been driven by mounting pressure to generate returns on AI investments. As spending on AI data centers and high-performance semiconductors has surged, investors have increasingly demanded measurable evidence of AI's impact on corporate cost structures. With capital expenditures, electricity consumption, and semiconductor procurement costs all rising simultaneously, companies have been forced to demonstrate not only AI's contribution to revenue growth but also its ability to reduce operating costs. Improving workforce efficiency quickly emerged as the most immediate solution. Geoffrey Hinton, Professor Emeritus at the University of Toronto, likewise told Bloomberg TV that beyond charging subscription fees for chatbots, the most direct path to monetizing AI investments lies in replacing human labor with lower-cost AI systems.

Enterprise Customers Focus on Lowering AI Costs

However, cost calculations for enterprise AI users have become increasingly complicated. As AI model usage rises, token expenses and cloud spending increase in tandem. Consequently, some corporate customers are reducing spending on OpenAI and Anthropic while restructuring costs by selecting lower-priced models based on task complexity. Uber, for example, announced this month that it had introduced tiered usage limits for certain AI tools, requiring employees seeking higher usage quotas to obtain separate approval. In April, Uber Chief Technology Officer Pravin Neppalli Naga disclosed that the company had exhausted its annual AI budget in just four months.

AI startup Lindy also shifted all of its AI traffic earlier this month from Anthropic's Claude model to DeepSeek's lower-cost open-weight model developed in China. Lindy CEO Flo Crivello said, "Immediately after the transition, we saw our cost curve fall straight to the floor." He added, "We have tremendous respect for Anthropic, but our company has been carrying AI costs that were unsustainable for far too long. This is simply a matter of corporate survival." He also noted that Lindy would consider returning to Claude if Anthropic reduced its pricing. Coinbase CEO Brian Armstrong recently outlined the company's AI cost management strategy as well, highlighting lower-cost foundation models, automatic routing based on task complexity, expanded caching, reduced unnecessary context, and greater transparency regarding AI usage costs. The objective is to maintain token consumption while lowering overall spending.

In response to these changes, AI developers are also adjusting their business models. Anthropic, in particular, has been refining its product strategy based on actual enterprise usage patterns. Its recently released Anthropic Economic Index analyzed more than four million Claude interactions using occupational data from the U.S. Department of Labor. According to the report, 57% of AI usage involved augmenting human work, while fully autonomous automation accounted for only 43%. AI adoption was especially prevalent in software development, document creation, and analytical work, but human review and decision-making remained integral to most real-world workflows.

Anthropic has also aligned its services with the ways enterprise customers deploy AI in practical work environments. Earlier this year, the company introduced plugin capabilities for its enterprise platform, Claude Cowork, enabling organizations to build dedicated AI agents for finance, marketing, customer support, data analysis, and other business functions. The initiative is designed to help enterprises integrate AI directly into existing operational systems. Scott White, who leads Anthropic's enterprise products, described the development as "a turning point where Claude evolves from being a simple assistant into a collaborative partner."

The strategy centered on workplace assistance is also evident in internal operations. According to The Wall Street Journal (WSJ), leading AI companies including Anthropic, OpenAI, and Google are deploying AI agents not only for drafting emails and summarizing meetings but also across finance, marketing, legal affairs, and customer support. Google, in particular, operates an AI system that reviews supplier invoices against contractual terms, reportedly increasing processing volume fivefold. Anthropic has likewise reorganized workflows so that AI handles repetitive tasks such as event planning and data entry while employees concentrate on review and decision-making.