More task focus, the rise of AI whisperers and improved observability -- AI predictions for 2025
Artificial intelligence has been one of the fastest growing areas in the tech sector over the past few years.
As AI becomes more commonly adopted what changes can we expect to see happening in 2025? Here are some expert views.
Avthar Sewrathan, AI product lead at Timescale, thinks we'll see greater consistency. "As AI apps become central to everyday interactions in 2025, consistency and reliability will take precedence. Engineers will face growing pressure to ensure their solutions provide accurate, user-friendly experiences that avoid misinformation or errors, safeguarding both user trust and brand reputation."
Dr Marc Warner, CEO of Faculty, partly echoes this, believing AI will move from experimental to essential:
The next wave of AI adoption will require a shift in perspective from senior leaders to stop viewing AI as experimental and to start treating it as essential to business transformation. The novelty of experimenting with tools like meeting summaries should not distract from much bigger gains that AI can offer in 2025 and beyond.
AI needs to be involved at the level of core business processes to achieve its full value. As a rule of thumb, AI use cases should be mapped to top level KPIs to have a meaningful impact.
Krishna Subramanian, co-founder and COO at Komprise, thinks governance processes will mature. "Protecting corporate data from leakage and misuse and preventing unwanted, erroneous results of AI are top of mind for executives today. A lack of agreed-upon standards, guidelines and regulations in North America is making the task more difficult. IT leaders can get started by using data management technology to get visibility on all their unstructured data across storage. This visibility is the starting point to understanding this growing volume of data better so that it can be governed and managed properly for AI. Data classification is another key step in AI data governance, and it involves enriching file metadata with tags to identify sensitive data that cannot be used in AI programs. Metadata enrichment is also available for aiding researchers and data scientists who need to quickly curate data sets for their projects by searching on keywords that identify file contents. With automated processes for data classification, IT can create workflows to continually send protected data sets to secure locations and, separately, send AI-ready data sets to object storage where it can be ingested by AI tools. Automated data workflow orchestration tools will be important for efficiently managing these tasks across petabyte-scale data estates. AI-ready unstructured data management solutions will also deliver a means to monitor workflows in progress and audit outcomes for risk."
Druva CTO, Stephen Manley, thinks ethics in AI will take a step forward in 2025. "Businesses will choose to lean into transparency and ethical AI to win over customers in an age of information chaos. In 2025, geopolitical turbulence will continue and misinformation is likely to abound. It's unlikely that new data privacy and AI policies will be passed and enforced in 2025, so customers will expect businesses to take responsibility for ethics in AI. As companies incorporate AI into their products, they have a responsibility to protect what and how the AI uses customer data, especially as it relates to sensitive data. Businesses must invest in ethical AI development, with an emphasis on transparency because AI adoption will directly correlate to the amount of trust the customers have in it."
Ville Somppi, senior vice president, industry solutions at M-Files, says, "In 2025, many companies will move beyond the exploratory phase of generative AI adoption, gaining a clearer understanding of where it can drive value and where it falls short. As large language models continue to evolve, becoming smarter and more affordable, organizations will experience improved consistency and quality in AI outputs. This affordability will open up new use cases previously unconsidered. Building on this, enterprises will increasingly focus on targeted, strategic AI use cases, refining their automation strategies. Successful organizations will recognize when AI can fully automate tasks and when it should support human work, shifting toward smarter, more practical applications that deliver measurable impact."
Bernd Greifeneder, CTO and founder of Dynatrace, sees a need for greater observability of AI services:
In the evolution of digital transformation, the rise of AI-based services introduces new complexities that make observability more critical than ever. With observability, teams will be able to build and operate new AI-powered digital services for performance and reliability, keeping cost, AI drift, user experience, and transparency in mind. These capabilities will give organizations the confidence to deploy AI technologies at scale.
As businesses deploy AI-driven services for predictive maintenance, financial forecasting, or cybersecurity, observability platforms will go beyond monitoring system performance to include visibility into AI queries. With this clarity, organizations can identify potential errors, correct biases, and ensure decisions align with both business goals and ethical standards.
In 2025, observability will play an even larger role, as organizations increasingly rely on AI to power critical services. By providing end-to-end visibility and actionable insights, observability will empower businesses to confidently scale AI systems, maintaining accountability, reducing risks, and building trust. Therefore, observability will no longer be an optional enhancement, but a mandatory component for delivering safe and effective AI-driven services.
Mona Ghadiri, senior director of product management at BlueVoyant, expects to see AI become more task based. "If you look at the history of software development, there is a sine wave of distributed systems that then go to monolithic systems and then back to distributed. With Artificial Intelligence (AI), we have been mostly thinking in monoliths -- companies deployed a chatbot to help their users (if they're lucky) or allowed their users to use things like BingChat. There was one place to ask, and it was trained in a general way. While useful, its lack of specific knowledge for industries, its biased towards the information its trained on, and lack of 'language comprehension' for AI the user level showed the AI industry that users rejected the general model for most of their jobs. I expect more distributed AI agents in embedded experiences that are narrowly specialized in discrete tasks. This would be parallel to allowing an AI car to drive, but only when the weather is nice and when there is no traffic. This has impacts both on operations and maintenance (more agents) and adoption of AI, as a smaller discrete task-based AI is easier operationally to implement the first one."
Stefan Weitz, co-founder and CEO at AI conference Humanx, predicts the emergence of a new specialist role. "A new class of high-paying roles will emerge for 'AI Whisperers' who specialize in fine-tuning and guiding AI systems in real-world applications. These experts will become as vital as engineers or product managers in AI-heavy industries. People in these roles will have a mix of technical chops, ethical awareness, super communication skills, and a non-linear mindset. Especially as AI jumps from vertical silos to cross-company systems, having people who can see across enterprises and align creativity with machine precision will be imperative."
CEO and co-founder of Entro Security, Itzik Alvas, thinks AI is set to improve identity and access management. "AI will play a significant role in the evolution of identity security, moving from traditional access controls to more dynamic and context-aware models. AI-driven systems will analyze behavior patterns to determine access rights in real time. For instance, AI will revoke access to an employee if it detects anomalous behavior, even if that employee's credentials haven't been stolen outright, based on real-time analysis of their actions."
Ken Dunham, cyber threat director at Qualys Threat Research Unit, also thinks AI will deliver security benefits. "AI-informed cybersecurity systems will become more adept at recognizing complex patterns and thwarting cyberattacks that would typically bypass human detection. At the same time, the prevalence of AI-driven attacks is expected to rise, with examples such as advanced phishing schemes and intelligent malware becoming more common. Federal agencies will continue to adopt continuous monitoring and adaptive learning technologies to counter AI-driven threats. These tools will evolve concurrently with the threat landscape, providing dynamic defenses capable of addressing sophisticated attacks."
Image credit: denisismagilov/depositphotos.com