Predictions 2025: Strategies to Realize the Promise of AI

Snowflake leaders offer insight on AI, open source and cybersecurity development — and the fundamental leadership skills required — in the years ahead.

As we come to the end of a calendar year, it’s natural to contemplate what the new year will hold for us. It’s an understatement to say that the future is very hard to predict, but it’s possible to both prepare for the likeliest outcomes and stay ready to adapt to the unexpected. 

In the enterprise technology space, both the greatest certainties and the most significant potential surprises come from one area: the rapidly advancing field of artificial intelligence. Thus, as we consider 2025 and beyond, it’s important to focus a lot of attention on the development and adoption of AI.

Together with a dozen experts and leaders at Snowflake, I have done exactly that, and today we debut the result: the “Snowflake Data + AI Predictions 2024” report. Along with issues of AI advancement, we considered directional trends and urgent needs in cybersecurity, open source software and more, but quite naturally a lot of our conversations turned to AI and how this fast-moving, volatile area of technology may continue to surprise the world.

2025 will be the year that many enterprises move from experimenting with LLMs and generative AI to operationalizing them, which will bring its own challenges. From my perspective, these are the key ideas that emerged from our discussions of AI and particularly its impact on the enterprise.

• AI observability is essential to operationalizing AI, and platforms will roll out solutions. When you’re running a large language model, you need observability into how the model may change as it ingests new data. It’s also important to have visibility into cost and performance. AI observability solutions are emerging to meet this need, but over time it’s most likely that the large data platforms, including Snowflake, will provide the solutions.

• Hallucinations will slow the rollout of customer-facing AI. The models keep getting better, and techniques such as retrieval augmented generation (RAG) will help reduce hallucinations and errors and put up guardrails that protect sensitive data and the voice and tone of a company. But businesses will continue to hesitate to put in front of customers a technology that may display bias or provide inaccurate responses. For this reason, internal-facing AI will continue to be the focus for the next couple of years.

• The next evolution in data is making it AI ready. For years, an essential tenet of digital transformation has been to make data accessible, to break down silos so that the enterprise can draw value from all of its data. This remains important, of course, but the next step will be to make sure that the enterprise’s unified data is AI ready, able to be plugged into existing agents and applications. 

  The trend to centralize data will accelerate, making sure that data is high-quality, accurate and well managed. Beyond working with well-structured data in a data warehouse, modern AI systems can use deep learning and natural language processing to work effectively with unstructured and semi-structured data in data lakes and lakehouses. Overall, data must be easily accessible to AI systems, with clear metadata management and a focus on relevance and timeliness. And data strategy must evolve to make sure that AI initiatives are aligned with business goals and are effectively instilling a data-driven culture in the organization.

• Expect autonomous agents, document digestion and AI as its own killer app. Our report notes that LLMs and generative AI will be so deeply embedded into how we live and work that thinking of a “killer app for AI” is like thinking of a killer app for electricity. But if we’re looking for the short-term winner, it’s going to be internal-facing use cases that let workers pull insights from massive troves of unstructured data. Snowflake recently helped a customer ingest about 700,000 pages of research and make it easily consumable through a conversational chatbot, allowing analysts to glean insights that had been functionally unavailable, though the company had the data. That will remain a major use of generative AI for some time.

  But in the next few years, the game-changing breakthrough in how we work with AI will be autonomous agents. Rather than answering a specific question, independent agents will act on broad instructions from a human user. “Create and launch a marketing campaign to attract this key customer cohort” could be automatically broken into subtasks such as designing on-brand copy graphics, making ad buys to reach the desired audience and optimizing based on initial performance.

• Leadership will be the antidote to AI exhaustion. AI has been advancing so quickly that the project that consumed a team’s every waking hour two weeks ago could be completely outdated tomorrow. Do you move forward or redo the work? If the latter, what if it happens again next week? Everyone I know in the AI space has talked about burnout at some point in the past year. To keep teams at their productive, creative best, leaders need to step in. We must set our sights on goals and ROI, rather than focusing on the shiny object. AI projects should not be about “the latest” or “the best.” Like any business decision or investment, we must weigh what’s most effective in terms of results and resources.

These thoughts are just some of what’s in the report. At the societal level, we look at the interplay of industry guardrails and regulatory oversight. Our cybersecurity experts tackle the ways that AI will both empower attackers and provide new ways to fight them. We look at developments in open source technologies that will allow organizations to improve their data strategies. And we talk about how leaders can keep up with the sometimes unnerving pace of change. Check out “Snowflake Data + AI Predictions 2025” for the whole story.