Several years since the first public-use generative AI tools, the adoption of enterprisewide AI strategy feels as inevitable as the adoption of the internet. What began as a simple chatbot can now design entire workflows, distill thousands of information sources into meaningful summaries, and even uncover novel insights well beyond any previous capabilities. Like all new tools, however, corporate-level AI usage comes with its risks, including a potential lack of personal accountability for decision making and tacit mistakes.

As large language models (LLMs) have been increasingly fine-tuned over the past year, though, their risks have appeared to attenuate. Among AI adoption advocates is Russ Wolfinger, Distinguished Research Fellow and Senior Director of Scientific Discovery and Genomics at JMP. He says that, especially in the last year, AI “is dramatically changing thought processes about what’s possible in terms of job roles, team dynamics, and the speed with which important tasks can be performed.” It is especially true in sectors like pharmaceutical development and semiconductor fabs, where AI can meaningfully change the day-to-day tasks of R&D staff.

Wolfinger exemplifies this paradigm shift. A lifelong programmer, he has embraced the capabilities of AI to write challenging new code efficiently. He explains, “I’ve loved hardcore coding and advanced analytics for over 40 years now, but the capabilities of AI engines are so impressive that I cannot help but switch mindsets.” Importantly, the use of AI models to help with challenging coding and analysis problems also democratizes programming. Wolfinger continues that “people who have never coded in their life are able to get these engines to do things they previously would have had to ask experienced analysts to do.” In this way, he says that “we are at an inflection point now where things are dramatically changing,” even just over the course of the past few months. Now, data analysis in science and engineering can be much more focused on which problems need to be solved – not the individual lines of code that need to be written.

Though Wolfinger is an enthusiast regarding the power of enterprise AI, he doesn’t believe all caution around it is misplaced. “Hallucinations are definitely real,” he warns, explaining that developing detailed, tailored contexts, skills, and prompts gives the highest probability that the model will stay on the correct track. High attention to detail, a watchful eye, and very rigorous testing, therefore, are major guardrails against AI tools going rogue. Further, Wolfinger argues that human involvement in AI-enabled processes is an indispensable second guardrail. He explains that “having humans in the loop is still absolutely vital, especially experts who can oversee what the engines do and validate their results.” In sectors where compliance is key, like many scientific R&D teams, having fundamental AI-awareness also boosts trust and helps keep auditors happy.

Efforts to create rigid, top-down rules for AI usage are logical, but Wolfinger argues that AI use cases are typically too broad to effectively collect one set of rules that always applies. Instead, he advises a “general attitude of cautiousness, but also an eagerness to push the edge for AI use and its superpowers.” Balancing ambition with caution is a good rule of thumb for any new AI use case, and using AI responsibly with smaller to increasingly larger and more difficult problems is the best way to proceed.

Wolfinger also continues to advocate for the role that AI plays best: an assistant. He says that even with all the amazing capabilities AI offers, people are forgetting that “your human core values, perspectives, and expertise are still the most important assets you have.” Therefore, having a corporate mindset not of replacing people with AI, but empowering them to take their work to a new level with AI, is a better path. In doing so, companies can avoid the fallout from poor AI-based decision making by ensuring that the models have the proper level of oversight at all times. The temptation to skirt personal accountability and hard work can also be quenched, as people are still in the driver’s seat with AI acting only as an advisor; they must continue to strive for excellence within the new paradigm.

A very important case study of human-centric AI strategy is in data analysis, which Wolfinger says LLMs aren’t directly proficient at yet. They are, however, very good at writing code and increasingly doing so with reasoning. What analysts can do is work with AI models to write efficient code and workflows for data analysis in a much faster way than ever before. Wolfinger says this strategy can “create a phenomenal collection of new advanced chat flows to accomplish just about whatever you want.” He explains that now, “we are making very good headway in analyzing data much more quickly, deeply, and effectively than ever before with assistance from LLMs.” Thus, AI enables a higher level of data analytics, helping to elevate analysts rather than replace them.

Regardless of the level of AI usage in each organization, Wolfinger believes that the benefits of AI adoption now clearly outweigh the risks. “In my mind,” he says, “this is one of the most radical, impressive improvements in tech that we have ever witnessed, and those who can best harness it in a human-friendly way will be those who win out.” By ensuring that people remain embedded in all workflows, and that AI augments rather than replaces human experts, effective and adaptable AI guardrails can be created that unleash its real advantages. Doing so is a must: regardless of industry or company size, he says “we cannot afford to ignore the capabilities of AI now. It’s an adapt or die mentality – AI evolution.”