2025. Generative AI trends that did not come to fruition (and why)
- Fully autonomous agents in mass production. Agents capable of performing tasks from start to finish were expected to be implemented on a large scale, but the reality was different. Companies failed to make the leap: according to McKinsey, nearly two-thirds did not progress beyond pilot projects. Only 39 per cent reported any financial impact on EBIT, and in most cases it was modest. Wharton reported that, although 82% of leaders used generative AI weekly and 72% measured ROI, it concluded that 2025 was a year of ‘responsible acceleration,’ focused more on improving productivity than on deploying autonomous agents across the board.
- End of hallucinations and perfect reasoning. Accuracy and reasoning issues remained. There were cases of misinformation even in advanced systems. OECD.AI warned of operational risks from incorrect responses and reinforced the need for frameworks to report risks. Stanford’s AI Index 2025 confirmed that, although there were advances, challenges in security and reasoning persisted, slowing down critical applications without additional controls.
- Regulation fully enforced across the ecosystem. The European AI Act came into force in 2024, but its implementation was gradual: bans and training began in February 2025, governance in August, and stricter requirements for high-risk systems will arrive on 2 August 2026. The full impact is expected in 2027. That is why, in 2025, many companies were preparing, not 100% compliant.
- Cost and hardware without barriers for all. Spending grew, but hardware took the lion’s share. Gartner estimated that $644 billion was invested in generative AI in 2025, with nearly 80% going to devices and servers. The idea of ‘software first’ was overshadowed by the need for AI infrastructure and equipment, driven more by supply than demand. IDC estimated £69.1 billion in generative AI alone by 2025, with strong growth through 2028, heavily dependent on storage and data.
- Uniform adoption and double-digit productivity for all. Adoption increased, but unevenly. According to the St. Louis Fed, overall adoption in the US reached 54.6% in August 2025, with only 5.7% of working hours performed with generative AI. The benefits were modest: Stanford’s AI Index reflected savings of less than 10% and revenue increases of less than 5% in most cases. Transformation, therefore, still needs time and maturity.
However, despite these unconfirmed auspices, we have not stopped using it, we have trained ourselves in it and we continue to follow its evolution both at a personal and corporate level.
So, what can we expect for 2026?
2026: Trends with high probability and relevance
Compliance with and implementation of the AI Act (EU)
From August 2026, all requirements for high-risk systems will come into force: risk management, data governance, transparency, human oversight, robustness and cybersecurity. The rules on transparency of generated content (Article 50) and practical guidelines will also be tightened.
What does this mean? Compliance, security and product teams will need to activate conformity assessments, labelling and traceability, as well as post-launch monitoring in their generative AI flows.
From pilots to measured value and disciplined scaling
Data from 2025 (Wharton and McKinsey) suggests that 2026 will be a key year: benchmarks and ROI measurement will be consolidated. Companies that redesign processes (rather than simply “coupling” models) and define clear growth and innovation objectives will achieve better results.
In a telco, this means applying agents in network operations, customer service and software development, always with robust controls and recovery systems.
‘Agentic’ architectures with governance and security by design
PwC predicts that agents will move from demos to useful tools, but with orchestration, metrics and responsible AI principles. The focus will be on controlled autonomy, multimodal capabilities and integration with data and APIs.
In telecommunications, proactive problem resolution, ticket management, deployment optimisation and field operations will be key.
Infrastructure: NPUs, GPUs, unified storage and data
The gap between expectations and reality in 2025 will drive investment in scalable storage, integrated data architectures, and model monitoring tools in 2026. All this to support inferences and adjustments in cloud and edge environments.
IDC anticipates a wave of high-performance infrastructure and unified data management to accelerate analytics and generative AI.
Market and skills: high adoption, but with critical skills
Although adoption will be widespread, differential results will occur where there is AI training, risk policies, and specialised talent. Deloitte and Wharton note that spending will continue to grow and that measuring ROI will be the norm.
Conclusions
2025 taught us an important lesson: generative AI did not deliver on all the promises we heard, just like any other developing technology. Autonomous agents did not conquer the business world, hallucinations remain a challenge, and European regulation is still under construction. Does this mean that the technology failed? Not at all. On the contrary, it has established itself as a strategic tool for improving productivity and efficiency, albeit with cautious steps.
Now, 2026 is shaping up to be the year of maturity. The AI Act will be a turning point: it will demand transparency, traceability, and real governance. Companies will have to move from scattered pilots to scaled projects with measured ROI and clear objectives. We will see smarter agents, but under control, secure architectures by design and a robust infrastructure — with NPUs, GPUs and storage ready to support the new wave of data.
But the real difference will not only be in technology, but in people. AI literacy, the creation of specialised roles and risk policies will be key to transforming generative AI into a competitive advantage.
In short: 2026 will not be the year of empty promises, but one of strategic action. Are you ready to take the leap? Telefónica accompanies you on this journey towards useful, secure and responsible AI.
