In 2025, smaller language models, larger context windows and, of course, agents will help enterprises embrace AI with more conviction.
Sentiment surrounding generative AI ebbs and flows but one constant remains: each exhalation of the news cycle breathes new life into the market. Even as 2024 winds down, much of the buzz surrounds innovation from GenAI pioneers such as OpenAI, whose large language models (LLMs) pace GenAI innovation.
However, as much as Open AI’s o1 model represented a leap forward for model reasoning there remains a disconnect between what the frontier models are pursuing and what most organizations—those in the 99th percentile—are seeking.
Even as OpenAI chases the Holy Grail that is artificial general intelligence (AGI), most businesses are leveraging AI to increase productivity while reducing costs and craft more compelling customer experiences. Some may also be using AI to pursue innovation and revenue streams.
To deliver these goals, more organizations are leaning more heavily on model inferencing, which maximizes where to run AI workloads based on performance, cost, data, security and latency. This marks the true implementation of enterprise AI.
And it’s a big reason why Menlo Ventures found that 72% of U.S. enterprise leaders expect broader adoption of GenAI tools across their organizations soon.
What does this signal for organizations in 2025? Trends that gained traction in 2024 will grow and mature in organizations. Here’s a sneak peek at what to expect in the year ahead:
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.
Is your organization ready to take that next leap forward?
Learn more about the Dell AI Factory.