AI Trends: What’s In, What’s Out, What’s Next
Artificial intelligence, more specifically technologies such as large language models (LLMs) and generative AI, have advanced rapidly in the past five years. Given the speed of change in these technologies and their associated industries, it can be difficult to identify trends and predict the next steps a company should take in this new environment.
Thankfully, Schneider Electric has grown used to staying ahead of the trends and pivoting with changing times during our 180+ years as a company. From becoming early adopters of electrical equipment to inventing the Programmable Logic Controller, Schneider Electric has been noted as a company that thrives in change.
So, what do we at Schneider Electric see as “in” and “out” in the trends of AI? Here are five of the most important transitions taking place in the world of AI today.
Artificial Intelligence and Operational Workflows
OUT: Focusing Only on Small Tasks
Much of the marketing for AI within business has centered on a replacement for the small, trivial tasks, such as writing a quick email or summarizing a meeting. These tasks support efficiency and accessibility, but despite their popularity, they reflect only the technology’s minimal capabilities. An informal study on how people utilize AI found that the top five uses for generative AI included brainstorming content ideas and utilizing generative AI as a search engine, which are relatively small tasks for AI.
IN: Utilizing AI as a Tool to Tackle the Most Complex Challenges
Creating a more sustainable world may be the most complex challenge humanity has ever faced. AI technology unlocks a whole new landscape of tools to solve the world’s most pressing challenges. The World Economic Forum’s report on AI’s potential in sustainability states that AI has had success in areas from agriculture to urban planning and recycling, and they argue that the power to solve sustainability issues with AI will only expand with the right guidelines in place.
Processes like decarbonization, scenario planning, supply chain engagement, and emission calculations all require mass amounts of data and constant surveillance of dozens, if not hundreds, of factors that could change the landscape at any moment. AI can help bring structure to complexity by continuously looking for significant, high-impact events, auditing and flagging anomalies in data, and modeling the effects of various strategies on a company’s carbon emissions and resource consumption. Schneider Electric is tackling these challenges through an AI-native ecosystem designed to transform how companies manage energy and achieve sustainability goals.
Enterprise Data Collection, Accuracy and Completion
OUT: Working with Random Data
Many AI models commonly in use today have been trained with data found on the internet from blogs, books, images, and more. As noted by an article in The Times on AI hallucinations, this free-for-all data environment not only poses risks of hallucinations from false information, such as adding glue to pizza sauce, but has also sparked court cases over copyright infringement, such as the case of Getty Images vs Stability AI. Low-quality data can be found everywhere, and if a model is using this data, the risks of mistakes and legal pitfalls rise substantially.
IN: Using Machine Learning to Verify and Audit Data
AI must manage a constant stream of information, and it must be able to swiftly and accurately resolve errors to provide trusted calculations and results. While AI models run, it can safeguard against anomalous results if it also performs self-audits along the way. In this scenario, the AI can flag any errors it believes it has made and share a degree of confidence along with its decisions. At Schneider Electric, we embed domain-expertise from our consultants and subject matter experts. This process ensures that the AI has expert-driven context and human oversight to improve the results our AI returns to our teams and users.
“We receive thousands of invoices from different regions, each using different units and formats. A human must memorize all these variations. That’s a huge cognitive load,” Schneider Electric AI-powered innovation leader Carlos Ribadeneira Espinoza explained in a recent article on the power of artificial intelligence and good data. “Many companies find manual data entry errors, typos, decimal misplacements—all of which can significantly impact downstream calculations. AI agents help catch those issues early. They can even flag missing or incomplete data and reduce the time lag in verification.”
Artificial Intelligence Training and Traceability
OUT: Closed-Door AI Training and Answers
Most companies using commercial AI disclose little about how their models are trained or how they generate results. This lack of disclosure leads to a “black-box effect” that reduces trust consumers have in AI results. Trust is eroded even further by “AI hallucinations", which are incorrect answers to prompts that the AI gives in confidence. A 2025 study on leading AI legal research tools found that, despite the companies’ promises of responses free from AI hallucinations, the models would hallucinate between 17% and 33% of the time. Despite the understanding that the answers given to them were false, the researchers could not understand why the specific hallucinations existed within the study, since the companies did not allow traceability or provide information about how the AI was trained. When deployed in industries where decisions are life-or-death, such as healthcare, or where mistakes could cost millions, such as in the legal system, the closed-door provision of answers has proven to be alarming.
IN: Built-in AI Traceability
We believe that AI performs better when we can trace the data and decisions from the model or agent. For example, users might wonder why an agent recommended solar over wind power, or vice versa. With built-in traceability, clients and consultants can determine the reasoning behind decisions and have a record ready for stakeholders, marketing claims, regulatory reporting, third-party auditors, or any other instance where the parties involved may need a record of what decisions were made and how.
Additionally, this built-in traceability will increase trust, as it will allow both users and developers to see whether there was an issue in the agent’s decision making and how that issue could be resolved in the future. With built-in AI traceability, every step toward the model’s decision, including data points referenced and actions taken, can be logged and reviewed by humans at any time.
Trust and Bias
OUT: Blind Trust in AI
When marketing AI, influencers in the private sector praise the technology and often avoid conversations about the limitations of AI. However, as found in a study on the views of artificial intelligence by Pew Research, the public is much less optimistic and are concerned about job loss, data misuse, and the loss of human connection.
As the public becomes more informed about AI hallucinations and other limits, consumers now expect businesses to be more forthcoming and transparent about the bias and limitations within their own AI models and agents.
IN: Understanding AI Bias and Limitations
At Schneider Electric, we see and understand the potential errors and impacts of using AI in an unresponsible way. We understand that the best use of AI in the private sector is with human oversight to balance out the inherent limitations of the technology. We acknowledge the environmental impact from increased AI usage and that AI models can make errors. We continuously work to minimize the impacts of these issues with human input, frugal use of AI, and supporting the decarbonization of data centers.
Use of AI
OUT: Over-use of AI
Funny cat photos. Digital dolls. Writing an email. These are all things that humans could create before the introduction of generative AI, but companies have marketed generative AI to do absolutely anything one could imagine. However, this diversity of applications comes at the expense of environmental degradation and water rations for towns near data centers worldwide. AI is an energy-hungry technology, and the energy demand for AI alone is projected to reach 44 gigawatts in 2025, which is over half of the total energy demand for all data center workloads. When we over-use AI, we dampen its potential to solve complex challenges like decarbonization and hurt those the technology is supposed to help.
IN: Frugal AI
Schneider Electric is committed to a vision of frugal AI, where we maximize the potential of these technologies while minimizing the negative impacts on communities and the environment. Through our understanding of economic, environmental, and resource pressures, we have created three rules for our use of AI: right-size every model, deploy workloads where carbon is lower, and build safety that drives efficiency. Not only will this vision help maximize efficiency and return on investment, but frugal AI practices help to contain the environmental impact of AI.
The I in AI: Intelligence
OUT: One Intelligence
In the past few years, there have been plenty of debates about the use of artificial intelligence in the private sector. Oftentimes, the participants in this debate have landed in one of two categories. The first believes AI will replace human intelligence due to its superior speed and rapid development. The other believes that the errors AI continues to make prove that we should completely forego using AI in favor of human intelligence. With these arguments, we limit the possibilities of what we can do based on one type of intelligence, which will ultimately fall to the weaknesses of that singular focus.
IN: Collaborative Intelligence
At Schneider Electric, we believe that there can be a world where the minds of humans work alongside artificial intelligence. Humans and AI each have tasks that they are individually skilled at, creativity and problem-solving for the former and maintaining databases and data auditing for the latter, and it is together that we will overcome our biggest challenges. We call this collaboration of human thinking and AI technology collaborative intelligence. Through collaborative intelligence, we can minimize the risks of both human and AI errors and break through the limitations of each independent intelligence to create meaningful change and progress in decarbonization. This approach will unlock the full potential of both AI and humanity, and it will allow for AI to power our future sustainably and responsibly.
Conclusion
As use of all types of AI, traditional, generative and agentic, continue to spread across industries, companies must not be satisfied with the status quo but rather continuously look to where AI can be used effectively and efficiently across their operations. By being ready to adapt to future technological innovation, companies can accelerate their growth and create more diverse opportunities for their portfolio. At Schneider Electric, we strive to continue our legacy of adaptability and our investment in agile, agentic AI systems, all of which position us to remain at the forefront of energy management, sustainability, and digital transformation.