The hype around generative AI in business has been relentless. From boardrooms to back offices, companies have been scrambling to harness the power of large language models to transform their operations. Yet, as 2025 unfolds, a sobering pattern is emerging: most enterprise AI efforts are falling flat.
A recent report from MIT’s NANDA initiative captures this gap, describing a “GenAI Divide” between the few organisations generating outsized returns and the many more struggling to move beyond pilots. But this isn’t just an MIT finding. It echoes a familiar story from the history of digital transformation.
A Familiar Pattern: Cloud, Mobile, and Now AI
When cloud computing first emerged, early adopters raced ahead while many incumbents hesitated, investing heavily in private builds that were costly and brittle. Similarly, the mobile revolution rewarded nimble startups while established firms were hamstrung by legacy systems. Generative AI is playing out in much the same way.
Startups can pick one narrow pain point, move fast, and scale. Enterprises, meanwhile, often attempt to shoehorn generic tools into complex workflows, or worse, build their own in-house systems. The result? Slow progress, high costs, and a lack of measurable impact.
The Real Barrier Isn’t the Model
Executives frequently cite regulation or model quality as the main barriers. But in truth, today’s models are powerful enough for a wide range of applications. The real challenge lies in organisational learning: how well tools adapt to workflows, and how well teams adapt to tools.
Generic AI assistants may delight individuals, but without contextual learning, they rarely create enterprise-level value. This is why automation in finance, HR, or supply chain—where outcomes are clearly measurable—often yields stronger returns than high-profile sales or marketing pilots.
Misplaced Bets on Building vs. Buying
History again offers a cautionary tale. Few companies today would dream of building their own ERP or CRM system from scratch. Yet many are attempting exactly that with generative AI—developing proprietary models and platforms in-house.
MIT’s data suggests that “buy and partner” strategies deliver success twice as often as “build it yourself” approaches. The lesson is simple: competitive advantage rarely comes from the tool itself, but from how intelligently it’s applied and integrated.
Workforce Shifts: Not Layoffs, But Non-Replacements
One of the quieter trends unfolding is workforce change. Despite fears of mass layoffs, what we’re seeing is attrition by design. Customer support, admin, and outsourced functions are shrinking—not through cuts, but by companies choosing not to refill positions once people leave. This mirrors the early days of automation in manufacturing: jobs didn’t disappear overnight, they faded gradually as technology proved cheaper and more reliable.
The Rise of Agentic AI
Looking forward, the next horizon is “agentic AI”—systems that don’t just respond, but act. Think of an AI that not only drafts an email but schedules the meeting, books the room, and prepares the agenda. These tools are already being piloted in advanced organisations and could mark the same kind of leap we saw when the internet moved from static pages to dynamic applications.
Bridging the Divide
For business leaders, the takeaway is clear: the GenAI Divide isn’t about technology, it’s about execution. The winners will be those who:
- Target specific pain points rather than generic use cases.
- Invest in partnerships with specialist vendors instead of chasing homegrown builds.
- Empower managers at the edge, not just centralised AI teams, to drive adoption.
- Focus on measurable value in operations and back-office functions, not just headline-grabbing pilots.
Generative AI is not a magic bullet. It is a powerful tool, but only when deployed with clarity, discipline, and realism. Just as cloud and mobile transformed business for those who embraced them strategically, AI will reward the organisations that cross the divide—not the ones who simply join the rush.
