(winyuu/Shutterstock)
Generative AI is the most promising technology of our generation, but it’s also proving difficult for organizations to use in an effective and profitable manner. That enigma makes GenAI rich ground for commentary and forecasts, which we hope is reflected in our second batch of 2025 GenAI predictions.
Large language models (LLMs) are, well, big. After all, it’s right there in the name. But in 2025, businesses will gravitate find there’s a lot of benefit to running small language models, says Hao Yang, the vice president of artificial intelligence at observability firm Splunk.
“Today’s LLMs know everything. But why do you need that? If you reduce the model to a reasonable size that fits your specific use case, you can reduce the cost significantly,” Yang says. “This is why we’ll see a rise in domain-specific small language models (SLMs), which will deliver unprecedented accuracy while significantly reducing operating costs and environmental impact. We know quality is important and domain-specific SLMs are far more accurate for any given use case, so they could help reduce the risk of AI hallucination and yield better results. We’re just beginning the next chapter of AI. The next generation of LLMs will be domain-specific.”
2025 will be the year that agentic AI jumps from demo-ware directly into the hype cycle, says Ciaran Dynes, the chief product officer at data integration provider Matillion, who also predicts that data governance will be back in vogue.
AI agents will deploy in 2025 (sdecoret/Shutterstock)
“Time will tell if the agents can be controlled and not go off into a never-ending cycle of handoffs between themselves but the early signs are very positive that agenticAI is about to kick off,” Dynes says. “It’s been a tough couple of years for data governance projects, however next year will see the reemergence of investment in data management and data governance practices. Data as product is the methodology of the day. But that’s a good thing for data teams.
Ariel Katz, the CEO of embedded analytics provider Sisense, isn’t one to argue semantics. But he does make a strong argument for the semantic layer becoming a key enabling tech for LLM adoption in the enterprise in 2025.
“In 2025, the semantic layer will become the crucial enabler for LLMs in enterprises, acting as a bridge between internal data and LLMs to deliver precise, contextually relevant insights,” Katz says. “By unifying enterprise data with global knowledge, this integration will revolutionize decision-making and productivity, making GenAI indispensable. Companies that embrace this convergence will dominate in innovation and customer experience, leaving competitors behind.”
The semantic layer–which functions as translator between how we speak about data and how computers store data–not only will play a bigger role in GenAI in 2025, but it will be instrumental for GenAI success, predicts Kaycee Lai, founder and chief strategy officer for data fabric provider Promethium.
The semantic layer figures heavily in GenAI development
“In 2025, while the workings of GenAI applications, copilots, and AI agents will be well understood by enterprises, their effectiveness and utility will hinge on the accuracy and relevance of the underlying data they leverage,” Lai says. “Achieving this accuracy will depend on having a robust semantic model that integrates directly with data, ensuring contextual understanding and relevance. These models provide a critical framework for aligning data with business terms, reducing risks of biased or misleading outputs, and improving the precision and trustworthiness of AI-driven insights.”
Search has been called the original AI use case, particularly with the neural search techniques that go beyond brute-force keyword matching. In 2025, the integration of GenAI and search will reach new levels, predicts Keri Rich, the vice president of product management and search provider Lucidworks.
“I expect to see different generative AI-powered experiences woven seamlessly into the entire commerce search and discovery experience next year, including saying goodbye to the search bar,” Rich says. “We’ve already seen plans from companies like Amazon to use Gen AI to expedite customer research and simplify product comparisons. Retailers first need to understand which parts of their shoppers’ journey could benefit from enhanced personalization and improved efficiency, and then develop the AI solutions to help them get there.”
Yash Sheth, COO & co-founder of GenAI observability company Galileo, comes bearing a veritable smorgasbord of GenAI predictions.
“1: GenAI will see an era of toolification, with AI technology becoming increasingly integrated with various software applications and systems for seamless integration;
“2: Enterprises will think deeply about domain and use cases as they balance broad capabilities and deep expertise. This will look like use cases with clear requirements, test sets, and evaluation criteria and a systematic approach to validation and performance assessment;
(Mirotvoric/Shutterstock)
“3: Multimodal AI will become a reality. There will be an expansion beyond text-only interactions, such as PDF document processing, insurance claim processing with mixed media and visual instruction processing (think IKEA furniture assembly);
“4: The industry has been led to believe that implementing AI into the enterprise is extremely easy. In 2025, we’ll see the realization that there isn’t an “easy button” and an evolution beyond simple interfaces (ChatGPT chatbot → RAG + Fine-Tuning) and into sophisticated implementations;
“And 5: Governance and compliance frameworks for more trustworthy AI will be at the forefront of 2025. Trust will shift from the blanket “trustworthy AI” claims to measurable reliability metrics such as specific, testable measures of model robustness, reliability, and known limitations, especially for edge cases and anomalies.”
When businesses started using GenAI technology two years ago, the focus was on using off-the-shelf LLMs trained on generic data. That worked for a while, but it won’t cut it in 2025, when private business data will rule the GenAI roost, says Matheus Dellagnelo, co-founder and CEO of Indicium.
“Today, however, models trained on generic data are no longer enough to deliver a competitive edge,” Dellagnelo says. “Businesses must also be able to connect models to their own business data so that the models can understand their unique business context and offer solutions tailored to it. Some companies may go a step further by training their own models using custom data, too–although that practice is likely to become common only for larger businesses with particularly complex and specialized AI needs. Either way, expect 2025 to be defined in part by efforts to connect models to business data in ways that weren’t important during earlier stages of AI adoption, when ‘off the shelf’ tools sufficed.”
Will GenAI learn to live within the guardrails in 2025? (phoelixDE/Shutterstock)
Most businesses have kept their GenAI projects on a short leash, lest the LLMs run amok and embarrass the company with erroneous output. But in 2025, GenAI will operate with greater independence and precision, particularly as advancements in symbolic reasoning and methods to combat hallucinations take hold, predicts Kelly Littlepage, CEO and co-founder of OneChronos, which provides search optimization for trading markets.
“In financial markets, AI adoption will progress incrementally across different functions. Back and middle office operations will see continued productivity gains through AI-driven human-in-the-loop (HITL) automation,” Littlepage says. “Trading desks are already leveraging generative AI to process semi-structured data more efficiently, though current capabilities are all HITL and stop short of completing trades or acting autonomously. As AI develops more sophisticated reasoning and explainability, it will gradually move into more complex front-office tasks and play increasingly central roles in trading workflows.”
GenAI tech is advancing at an incredible rate, but humans typically are still required to update foundation models. In 2025, the models will begin to update themselves, predicts Mike Bachman, the head of architecture and AI strategy for data integration and automation firm Boomi.
“Memory improvements and techniques associated with retrieving and merging greater levels of context pave the way for LLMs to adaptively learn, improve their answers, and update their own ‘world model,’ once deployed,” Bachman says. “Expanded context windows, quantization techniques, and agent-based information retrieval will allow a model to intrinsically update itself without retraining. As enterprises invest more resources in data quality up front, LLMs with better memory systems will be able to update their foundational training with novel information based on more context-dependent, filtered, and accurate information — and it’s only a matter of time before they’re so well-informed that they’re able to write, debug, and improve themselves.”
Quantum computing (QC) has made a lot of progress in recent years, to the point where companies are close to finding real-world applications for QC. In 2025, the combination of QC and AI will start to look intriguing for business, says Enrique Lizaso Olmos, the CEO and co-founder of Multiverse Computing.
The junction of QC and AI is a fertile area for development (Dave Hoeek/Shutterstock)
“In 2025, Quantum Computing will further solidify its position as a transformative technology with real-world applications. Also, the synergy between quantum computing and artificial intelligence (AI) will become increasingly evident. Quantum technology is emerging as a critical tool for enhancing AI’s efficiency, while AI plays a key role in integrating quantum solutions into practical applications. This reciprocal relationship has enabled both technologies to address their respective challenges more effectively.”
Lots of companies have lots of data just lying around and not providing any value for organizations. In 2025, that dynamic will change, as GenAI turns data graveyards into AI goldmines, predicts Haseeb Budhani, co-founder and CEO of Kubernetes automation firm Rafay Systems.
“Organizations are sitting on ‘data graveyards’– repositories of historical information that became too resource-intensive to maintain or analyze. This is largely because it can be expensive to tag data and keep track of it,” Budhani says. “Many companies defaulted to ‘store everything, analyze little’ approaches due to the complexity and high costs related to data management. Yet valuable insights remain buried in emails, documents, customer interactions and operational data from years past.
“With GenAI tooling, there’s an opportunity to efficiently process and analyze unstructured data at unprecedented scale,” Budhani continues. “Organizations can uncover historical trends, customer behaviors and business patterns that were too complex to analyze before. Previously unusable unstructured data will become a valuable asset for training domain-specific AI models.”
GenAI tech is impressive, but it requires a considerable amount of human skill to build successful GenAI apps. In 2025, organizations that prioritize AI literacy will gain a competitive edge over those who don’t, predicts Eric Sydell, the CEO of Vero AI, which builds AI monitoring tools.
“As AI becomes increasingly integrated into work processes, the digital divide is set to widen. Those who learn about and leverage advanced tools, particularly generative models, will gain a significant productivity edge, driving career growth and business productivity,” Sydell says. “This shift will create a rising demand for AI literacy and upskilling programs. Although AI will automate routine tasks, driving organizational changes and displacing some roles, it will also create new opportunities to manage and oversee AI systems. The need for AI governance and human oversight will intensify, requiring businesses to balance machine efficiency with human expertise.”
Related Items:
The Top 2025 Generative AI Predictions: Part 1
2025 Big Data Management Predictions
2025 Data Analytics Predictions
Source link
lol
