Current limitations and future solutions: what can we expect from AI in 2026?

—

© [2025] [𝘒𝘐-𝘎𝘦𝘯.] 𝘔𝘢𝘳𝘤𝘦𝘭 𝘖𝘩𝘳𝘦𝘯𝘴𝘤𝘩𝘢𝘭𝘭. 𝘉𝘢𝘴𝘪𝘦𝘳𝘦𝘯𝘥 𝘢𝘶𝘧 𝘞𝘦𝘳𝘬𝘦𝘯 𝘷𝘰𝘯 𝘈𝘯𝘥𝘳𝘦𝘢𝘴 𝘖𝘩𝘳𝘦𝘯𝘴𝘤𝘩𝘢𝘭𝘭.

According to Microsoft CEO Satya Nadella and OpenAI, a “model overhang” or “capability overhang” is preventing AI models from delivering the level of performance they are theoretically capable of. What does this mean? AI models have enormous potential, but much of it often remains untapped because users lack the skills needed to make full use of their capabilities. The organisations and individuals cited above therefore argue that the way people interact with AI models needs to change: away from simple, isolated prompts and towards complex workflows involving multiple AI models and agents integrated into existing systems.

This view is far from uncontroversial. Experts have criticised it as too one-sided, while critical responses on social media increasingly argue that it shifts the blame for poor results onto users’ supposedly inadequate behaviour. But could the shortcomings also be rooted in product design? Some users say that working with AI tools feels like using “slot machines”: they encourage users to keep going even when it is already clear that numerous rounds of correction may be needed to reach the desired answer. Some report requiring as many as eight prompts. Others are already describing a new phenomenon of “prompt fatigue”.

What could follow from this in 2026? According to the IBM Trend Report 2026, there is broad agreement that the future belongs to multimodal AI systems. This would also require the human role to shift away from operational tasks and towards strategic orchestration. Professionals in areas such as marketing or programming could increasingly become AI orchestrators, combining different autonomous tools and agents into complex end-to-end solutions. Human intuition and strategic direction would therefore move to the centre of working with autonomous systems.

Whether these trends materialise and what they mean for the use of AI in research is something we will continue to monitor in this newsletter over the coming weeks and months.

By Marcel Ohrenschall