Sam Altman, CEO of OpenAI, stated that 2025 would be the year AI agents begin operating at full capacity. Many companies adopted this vision. AI tools appeared in Google search, Microsoft Office applications, and live chat systems across numerous services. Organizations began replacing programmers with AI coding systems. Entire departments faced layoffs across many countries. These reductions occurred primarily in late 2024 and early 2025. Despite the expectations, the prediction has not materialized.
Research shows that even the most advanced AI agent, developed by Anthropic, completed only 24% of assigned tasks. Operational costs, including server and maintenance expenses, exceeded initial estimates.
Advertisеment
A Gartner survey reveals that over half of CEOs now plan to abandon efforts to significantly reduce customer support staff by 2027. This applies to customer support roles, where tasks are typically routine.
Public relations teams are adjusting their messaging. Phrases like “hybrid approach” and “transition challenges” are replacing claims of AI surpassing human automation. The need for human oversight in managing workflows remains significant.
Katie Ross, senior director of customer experience insights at Gartner, stated that the human element remains essential in many interactions, and organizations need to balance technology with human empathy.
Workers See Through the Hype
A report by GoTo and Workplace Intelligence shows that 62% of office workers believe AI hype is “vastly exaggerated.” Most employees use only free tools like ChatGPT and say they would not use paid AI assistants. AI skepticism among executives stands at 49%, but this figure has risen over the past six months despite new releases such as GPT-4.5.
Case Studies: Reversals Across Industries
Klarna
The Swedish fintech company reduced its workforce by 22% in 2024, laying off 700 employees to replace them with AI assistants. It expected annual savings of $40 million. By May 2025, Klarna reversed course and launched a mass hiring campaign to rehire former employees. Service quality had declined significantly. The cycle of layoffs, severance payments averaging $8,160 per employee for four months, and retraining cost at least $15 million. The projected savings did not occur.
IBM
Between 2023 and 2024, IBM automated parts of its HR department using a virtual agent called “AskHR” to handle requests, documents, and vacation processing. The company laid off approximately 8,000 employees. AI could not manage tasks requiring empathy, subjective judgment, or personal interaction. IBM resumed hiring for thousands of positions.
CEO Arvind Krishna confirmed that overall employment grew due to reinvestment in areas where AI cannot replace humans: “While we've done a lot of work internally at IBM to use AI and automation in certain enterprise workflows, our overall employment has actually grown. We've been investing more in other areas, so it was inevitable.”
Duolingo
The educational platform reduced freelance writer and translator contracts in late 2023 and 2024. By 2024, it cut 10% of freelance roles, with plans to reduce up to 90% of its content team by year-end. In 2025, the company halted further layoffs. Users criticized AI-generated courses as “template-based,” “boring,” and lacking cultural nuances. Public backlash grew on social media. Duolingo, once a leading advocate of AI-driven layoffs, now distances itself from that stance and has resumed hiring for previously eliminated roles.
Similar reversals occurred at Chegg, Dropbox, SAP, Google, Zoom, BuzzFeed, and BT Group. In each case, attempts to replace entire departments with AI led to operational costs and forced rollbacks.
AI Slows Down Experienced Programmers
A 2025 study by the METR institute found that AI tools slow down experienced software developers. The research observed 16 experienced open-source developers solving 246 real-world tasks, including bug fixes and feature implementation, in large codebases. Participants used tools such as Cursor Pro with Claude 3.5 or 3.7 Sonnet, or GPT-4.5.
Developers expected AI to speed up their work by 24%. After completing tasks, they believed productivity improved by about 20%. Actual results showed a 19% increase in task completion time when using AI.
Researchers identified several reasons.
- Developers’ optimism exceeded AI capabilities.
- Familiarity with codebases reduced opportunities for AI optimization.
- Large, complex projects (over one million lines of code) challenged AI performance.
- AI-generated code required extensive review; developers accepted less than 44% of suggestions.
- Up to 20% of time went to checking and debugging AI output.
- AI struggled with implicit context in large repositories, leading to irrelevant suggestions.
The study used rigorous methodology. Developers estimated task duration with and without AI, recorded screen activity, and reported time spent. Participants received $150 per hour for accurate performance. Results remained consistent across all subjects.
Researchers caution against broad generalizations. The findings apply to experienced developers working on complex systems. AI may benefit less experienced programmers or those handling small, unfamiliar projects—roles typically lower in cost.
Despite the slowdown, many developers plan to continue using AI tools. They report that while AI does not accelerate work, it reduces monotony in repetitive tasks.
AI performs well on routine, template-based code, code analysis, answering questions, and performance suggestions. It struggles with novel applications, uncommon languages, and complex frameworks.
Meta’s AI Spending Signals a Trend Peak?
Meta, formerly Facebook, has a history of entering trends late. It heavily invested in the metaverse, rebranding from Facebook to Meta in 2021. The company spent $47 billion on the initiative. Daily users of its “Horizon Worlds” platform dropped from 900 to fewer than 300.
In 2024, Meta shifted focus to artificial intelligence. It laid off employees from its Reality Labs division and launched a “superintelligence” lab. The company is poaching top AI talent from OpenAI. They are paid much more for this transition than Cristiano Ronaldo: up to $300 million over four years, including more than $100 million in the first year - mostly in shares with immediate vesting. Meta has recruited eight such specialists, spending over $2 billion.
Meta plans to spend $70 billion on AI data centers in 2025, double the previous year. This includes investments in custom MTIA and MSVP ASIC chips, the RSC supercomputer, and the $14 billion acquisition of Scale AI.
Unlike Google and Microsoft, Meta lacks a clear path to monetize its AI developments. So far, Nvidia remains the only public company to achieve significant financial gains from the AI boom, supplying essential hardware.
AI will not repeat the metaverse story exactly. However, Meta’s massive investment may indicate that the peak of the AI trend has passed. The company’s tendency to join trends at their end raises questions about the sustainability of current AI enthusiasm.
Support us
Winaero greatly relies on your support. You can help the site keep bringing you interesting and useful content and software by using these options:
