Now that 2024 has come to a close, it’s time to look ahead and think about what may be coming in 2025 in the world of AI and IT operations. These are my top 5 predictions of AI trends for the coming year:
1. Enterprises will need fine-tuned large language models to get useful value from AI
Large language models (LLMs) are powerful, but their reliance on massive datasets presents challenges. These datasets often contain biases and lack up-to-date information specific to individual enterprises.
To address this, there will be a trend of enterprises moving towards two key solutions:
Specialized LLMs from vendors
Instead of relying solely on massive, general-purpose LLMs, organizations will increasingly adopt smaller, more specialized models. For enterprises to get the desired value out of vendor-provided AI integrated with their SaaS solutions, they will need to collaborate with the vendor on a fine-tuned or partially pre-trained model, with training tokens or data that are unique to the enterprise. In 2025, it will become much cheaper to pre-train and fine-tune models via better algorithms and data efficiency.
Reinforcement learning from human feedback
Reinforcement learning from human feedback (RLHF) allows users to refine LLMs by incorporating feedback from people, effectively "filling in the gaps" in their knowledge and mitigating biases that are present in the underlying training material. This is particularly valuable when dealing with proprietary or rapidly changing information within an organization where training tokens are limited or hard to come by. SaaS vendors providing AI value will incorporate RLHF into their offerings as part of standard workflows.
2. More transparency into the AI “black box”
One of the primary concerns during the initial phase of AI adoption was the difficulty in understanding and explaining how AI systems reached their conclusions. More recently, there has been a greater focus on teams rapidly building pilots to test what they can achieve with AI, with some providers deliberately hiding their workings as their intellectual property. However, I think the industry will change as more users demand the ability to see how the AI “thinks” and citations of the data that was considered. This does have the advantage that, in collaboration with the vendor partner, the enterprise can build excellent training data for their own purposes.
3. For TechOps functions to see the desired value from agents, they need to move away from being solely text based
At the moment, many AI agents are being deployed as advanced chatbots that can interact with existing knowledge bases, customer relationship management software, and support ticketing systems. There is some value here in the advanced chatbot approach. For agents to be more valuable in 2025, providers will expand the ways that AI chatbots interact with users, not just in text form but across user interfaces and visualizations of important processes, workflows and data.
4. AI will also be used to enhance cloud modernization
Cloud migrations are becoming increasingly sophisticated and complex. While "lift and shift" remains a common approach, there's a growing trend toward application modernization. In 2025, cloud modernization will involve leveraging AI-powered development tools to automate replacing legacy components with cloud-native and resilient architectures.
5. An increase in automation and AI won’t eliminate the need for people
Although there has been a lot of excitement in recent years about the benefits of AI and automation, there is also fear around these newer technologies eliminating the need for people in the workplace. However, the main goal of innovations in AI and automation is to empower and augment people to be more productive and get better outcomes. We see people being in the loop for IT operations in major organizations continuing to be highly important, even as automation increases. Even with agentic AI performing complex problem-solving tasks, people will still need to set the parameters of how those systems work and make key decisions, with the help of real-time data, to ensure that requirements are being met.
Expect to see more multi-modal AI systems that incorporate human oversight to ensure accuracy and reliability. This "human-in-the-loop" approach will become increasingly prevalent.
2025 is the year where the join between data, machine learning, and orchestration will give organizations the ability to achieve executable IT DR
Between 2023 and 2024, the number of outages organizations faced increased and they took longer to recover from. We see this trend continuing, especially for organizations that have not invested enough in automation or AI in disaster recovery. Our survey of 300 technology leaders found that 82% recognized they need to increase investment in IT disaster recovery software automation.
It has been very hard for organizations to use their IT DR capability in real major incidents as often the plans are not executable, not connected via API to be automatically configured and instantiated, not connected via API to execute the failover, and don’t lay down the training tokens for future learning, and they can’t create enough of these basic plans to cover every eventuality.
Current levels of AI, data, and orchestration will change that in 2025. We will see organizations adopt stores of application data that AI can parse at speed to determine the recovery scenario and they will adopt recovery plans in the form of AI-enabled runbooks that can orchestrate across humans and machines to give viable options to incident managers to be able to trust IT DR in real recovery scenarios. This is Next Generation IT DR and 2025 is the year you can make it happen.
Find out more about what the next generation of IT disaster recovery will look like in 2025.
Do you agree with my predictions? What are your predictions for IT operations in 2025? Let us know on LinkedIn or X!