The 2025 AI Trends Turbocharging The Enterprise

The perception of generative AI continues to ebb and flow, but one thing remains unchanged: each surge in the news cycle breathes fresh momentum into the market. As 2024 draws to a close, much of the attention still revolves around innovations from GenAI leaders like OpenAI, whose large language models (LLMs) set the pace for advancements in the field.

Yet, while OpenAI’s o1 model marked a significant step forward in model reasoning, a disconnect persists between the ambitions of cutting-edge models and the needs of most organizations—the majority operating within the 99th percentile.

While OpenAI pursues the elusive goal of artificial general intelligence (AGI), most businesses are leveraging AI to streamline operations, boost productivity, cut costs, and create more engaging customer experiences. Some are also tapping AI to fuel innovation and unlock new revenue opportunities.

To achieve these objectives, a growing number of organizations are focusing on model inferencing—optimizing where and how AI workloads run based on factors like performance, cost, data privacy, security, and latency. This approach represents the practical and scalable application of enterprise AI. This shift is a major reason Menlo Ventures reports that 72% of enterprise leaders in the U.S. anticipate a wider adoption of GenAI tools across their companies in the near future.

 What does this mean for organizations heading into 2025? The trends gaining momentum in 2024 will continue to evolve and solidify, shaping the business landscape even further. Here’s a glimpse at what lies ahead for the coming year.

Big things come in small models. Small language models refined with retrieval-augmented generation (RAG) became the go-to choice for most organizations deploying GenAI in 2024. In 2025, these models will become standard practice across organizations looking to capitalize on GenAI opportunities without ceding control over corporate data or busting their budgets. This will be critical in accelerating GenAI deployment from on-premises systems out to the edge, where small models will run on AI PCs at a fraction of the deployment cost with minimal latency. “SLMs unlock new categories of AI features and offline functionalities at the edge, which requires less expensive hardware, connectivity, and energy to operate,” noted Jim Rowan, Deloitte’s head of AI.

Large Context Windows Catapult AI Performance. While smaller models are certainly popular, AI model makers are also boosting their models’ short-term memory, or context windows, essentially the number of words they can process. A business’ ability to maintain coherent prompts is critical to helping users generate the answers they seek from prompts. Google, for example, increased the context window of its NotebookLM research assistant from 1,500 to more than 1.5 million words in less than two years. Businesses will also benefit from leveraging large context windows to comb through their entire corpus of documents—aka institutional knowledge—in a single prompt. For most businesses, the context window limitation may well be an afterthought by the end of 2025.

Wanted: AI-flavored Soft Skills. Educating users on prompting for almost every business function gained momentum in 2024. Expect a more concerted effort to help individual employees explain how they use AI to do their work to colleagues in 2025. What might this education track look like? New and improved soft skills. Imagine an engineer explaining their chain-of-thought prompt process and resulting outputs to stakeholders across business lines. If GenAI democratizes how workers use AI in their roles, sharing the wisdom gleaned from the execution and practice is the next practical step. And employees seem generally open to this role, with 76% of workers surveyed by Slack saying they want to become an AI expert.

Agents Will Go Forth and Multiply. Well, maybe. Autonomous software bots that make decisions were the talk of GenAI tech in 2024, tempered by the position that agents need stronger reasoning capabilities to add value. The fits and starts will continue in 2025, however, you will see broader adoption among progressive enterprises committed to tackling the challenges of agentic AI architectures, which include striking the right balance of short-term and long-term memories. Organizations will ramp up adoption of agents, which will incorporate multimodal capabilities and leverage RAG, facilitating greater enterprise intelligence. As Madrona Venture Labs partner Jon Turow said: “Imagination, user trust and intuitive design will separate the winners — teams will adopt agents that feel natural and prove their reliability in real workflows.”

Governance for Good AI? More, please! The need for governance and oversight into the digital black boxes that describe most AI models has never been greater. Yet there is a surprising lack of oversight at the board level, with 45% of board members saying that AI has not made it onto their agendas at all, according to Deloitte research. In 2025, AI will absolutely be on more boards’ radars, driven partly by broader corporate adoption but mostly by fears about reputational risk. “We’re at a point where responsible AI use can reshape industries and drive growth of organizations that embrace it,” Deloitte’s Rowan said. “But without governance and oversight, even the best tech can fall short.”

The Bottom Line

Smaller language models, larger context windows, education and soft skills, agents and governance are just a handful of the trends that will impact enterprises in 2025. Expect more AI trends including some that haven’t yet presented themselves—to unfold as the year progresses.

In the meantime, organizations must continue to test and learn from their GenAI deployments and pass those learnings on to employees across business lines. This includes the ability to deploy AI responsibly for employees and customers.

No organization should embark on their GenAI journey alone. Trusted advisors can help organizations execute on their AI strategies, counseling on anything from the selection and deployment of AI-enhanced technologies and to modern frameworks.