Credit: roberthyrons / iStock
Generative artificial intelligence is now in its third year of widespread adoption, bringing a mix of massive opportunities and big challenges that are redefining industries.
It has the potential to transform enterprise performance — streamlining workflows, reducing costs, increasing productivity, and enabling more intelligent, data-driven decisions. But there are significant obstacles, particularly around scalability, data readiness, and workforce alignment.
Many organizations are struggling to modernize their infrastructure and integrate generative AI into legacy systems while ensuring that their teams are prepared to work alongside these advanced tools. If these foundational hurdles aren’t addressed, the full potential of generative AI may be difficult to realize.
“The journey starts with how we build our modern ‘digital core,’ and that will be the immediate focus because otherwise, if we’re just jumping around the [generative] AI use cases, we’re not going to go too far,” said Rose Mei, vice president of integrated business planning and AI at chemicals company Tronox.
Mei joined industry experts from Salesforce, Corning, and S&P Global for a panel discussion at the MIT Initiative on the Digital Economy’s Business Implications of Generative AI Conference. The panel was moderated by Lan Guan, chief AI officer at Accenture.
While there is no silver bullet, experts shared six elements of a successful strategy for harnessing generative AI to amplify and support human workflows.
Focus on augmentation
The panelists underlined the importance of using generative AI to enhance and complement human workflows (augmentation) rather than replace them (automation).
“We’re beyond the generative phase — we’re studying the autonomous enterprise,” said Vala Afshar, chief digital evangelist at Salesforce.
He raised the subject of “agentic AI,” where digital “employees” go beyond creating content to actually taking action alongside their human counterparts. The goal is not just to cut costs but to tackle inefficiencies and reduce waste.
“We’re now at a point where we’re not building software that can help you do the work — we’re actually building software that does the work alongside you,” Afshar said, estimating that in a year’s time, Salesforce will have a billion AI “agents.”
Keep humans in the loop
Ensuring worker safety and efficiency is critical, especially in labor-intensive industries like manufacturing. At materials science company Corning, this means grounding generative AI with historical maintenance logs and real-time production data so that operators can make quicker, more informed decisions while reinforcing safety and reliability on the production floor.
“You are seeing technology that’s being used by a manufacturing operator to do the operations more seamlessly, with better quality and better reliability,” said Soumya Seetharam, senior vice president and chief digital and information officer at Corning.
Businesses should also look beyond business outcomes, tackling questions about AI’s impact on society.
“What are humans for?” asked an audience member, raising concerns about how generative AI could displace workers.
Panelists noted the need for businesses to be transparent and build trust. “We need a culture shift in how we think about AI and the value of humans,” said Savannah Thais, an associate research scientist at Columbia University’s Data Science Institute, stressing the need for enterprises to avoid the urge to automate art, creativity, science, and inquiry.
Modernize the data infrastructure
A central theme was the critical role of robust data infrastructure as the foundation for scaling generative AI.
“Every AI project begins and ends, or lives, as a data project, so infrastructure assessment … is key to your success,” Afshar said. He highlighted the challenges enterprises face due to legacy systems and fragmented, siloed data.
“Only 4% of enterprises have data ready to be ingested” by AI models, he noted.
A key challenge: Much of the data that companies rely on is stored in structured formats, like tables and databases. However, today’s AI models struggle to retrieve information from such systems effectively, particularly when it involves using SQL, the programming language designed to query databases.
“I do think that connecting your data to large language models, to these generative AI applications, is going to be the difference,” said Bhavesh Dayalji, chief AI officer at S&P Global and CEO of Kensho Technologies. “You’re not going to always be able to use this pretrained model to then put into production some solution that customers are going to use.”
The bottom line: Investing in robust, interconnected data systems is an essential first step for scaling generative AI.
Upskill and educate talent
The panelists stressed the critical human dimension of integrating AI into enterprises. This involves equipping employees with the right skills, tools, and mindset to collaborate effectively with AI while also addressing the broader societal impacts. Leaders’ priorities should include mitigating job displacement, fostering trust in AI systems, and ensuring that the technology benefits both the workforce and the wider community, they said.
“You need oversight by a human. You can’t de facto trust what the machine tells you,” Seetharam said.
The panelists detailed efforts to educate employees about AI’s role and provide them with tools to enhance their work. They agreed that empathy and curiosity will be essential skills for the workforce of the future.
“When you’re hiring, look for people with curiosity, healthy skepticism, and a learning mindset,” Seetharam said.
Related Articles
“The hottest new programming language now is English,” said Afshar, highlighting AI’s ability to automate coding and the growing importance of natural language interfaces. These are systems that allow users to interact with technology using everyday language, making it more intuitive and accessible.
Understand limitations
Businesses need to be aware of the drawbacks of generative AI models to avoid common pitfalls, including “hallucinations,” where they generate information that seems plausible but is entirely false. Also, they often struggle to generalize effectively across different contexts or datasets.
“These are not causal engines, still; they’re not designed to be. They are inherently probabilistic, and so they are going to continue to hallucinate, and you cannot expect them to perfectly model your problem,” Thais said.
To mitigate these risks, companies should invest in training employees to critically evaluate AI models, understand their inherent biases and constraints, and recognize where they are most likely to fail, the speakers said.
Make big bets
The panelists warned that although generative AI is opening doors to countless possibilities, businesses need to narrow their focus. Rather than chasing every shiny opportunity, they must zero in on a few high-impact initiatives — projects that not only deliver measurable value but also align strategically with long-term goals.
“At some point, you’re going to have to [act],” Dayalji said. “I think we’re at the phase now where we’ve learned enough, and I think it’s now about putting your money where your mouth is.”
AI Executive Academy
In person at MIT Sloan
Register Now
