loading
loading
While the West debates slowing frontier AI development, enterprises are rushing to adopt the technology across their business processes. But deploying AI at scale remains a challenge, as enterprises struggle to keep pace with rapidly advancing capabilities. Industry leaders across AI, data science, financial services, and technology say the challenge is no longer limited to model performance. Organisations must also address unreliable data, integration with existing systems, governance, implementation costs and accountability for outcomes. For some, the industry is moving beyond the hype surrounding new models. Others see the current phase as an opportunity to establish safeguards before AI systems take on more consequential business decisions. Pilot to production Ram Kumar Nimmakayala, Product Leader (AI/ML & Data) at Western Governors University in the US, argues that the perceived slowdown reflects a change in priorities rather than a decline in AI adoption. “Honestly? From where I sit, there’s no slowdown — there's just less hype to hide behind. I spend my days inside the organisation trying to get AI to actually work, and the model was never the hard part,” he says. The bigger challenges are poor data quality, resistance to changing existing workflows, and a lack of clear ownership. According to Nimmakayala, nine out of ten AI pilots fail to reach production, rarely because of the model itself. He therefore disagrees with the narrative that AI adoption is slowing down. “The game just moved from ‘look what the model can do’ to ‘can you make it stick inside a real business’ — and that's a much harder game. AI governance and enablement are getting more critical. It is all about AI Value and RoI now,” he adds. Rakesh Pal, an insurance industry leader driving AI adoption in India’s financial services sector, agrees with the assessment, emphasising the need to assess organisational readiness before moving forward with AI initiatives. “I strongly believe that many organisations need to take a step back to work on the readiness parameters,” he says. Even when a model performs well, businesses must establish the data infrastructure, workflows and ownership needed to deploy it effectively. Governance could determine AI adoption Guru Ananthanarayanan, Managing Director for India at Cognite, views the slowdown as an opportunity for the industry to establish stronger governance and responsible AI frameworks. “I see a slowdown as a ‘rest before the sprint.’ It gives the AI industry time to put the right governance and responsible AI frameworks in place, especially for frontier LLMs, so what we're building stays within those guardrails,” he says. He believes that proper governance and safeguards will let the industry advance with greater confidence, shifting the debate from humans versus AI to how people can use AI responsibly. Dr Prashant Ramappa, Chief (Principal) Data Science, AI, Generative AI and ML Leadership at AT&T, distinguishes between the pressure on frontier model developers and the pace of enterprise implementation. “I don't buy the slowdown narrative—what we are actually seeing is exhaustion at the model-creation layer. Frontier AI companies sparked a relentless weekly arms race of leapfrogging benchmarks, and now they are hitting a tipping point of team burnout and massive capital burn from endless cycles of training, failing, and retraining,” he says. He argues the slowdown isn't happening at the enterprise adoption level. Instead, AI model developers need more time to establish governance frameworks, strengthen safeguards, fix security vulnerabilities and put human oversight mechanisms in place. “They need to slow down to make these systems safe, but because they are trapped in a race of their own making, nobody is actually taking their foot off the gas,” Ramappa says. His assessment highlights the tension between the competition to develop more capable models and the work required to make those systems safe and dependable. The enterprise autonomy gap Dr Rajan Gupta, Enterprise AI & Data Technology Leader at Deutsche Telekom Digital Labs, argues that AI capabilities are progressing faster than enterprises can absorb them. He describes this disconnect as the ‘Enterprise Autonomy Gap’. “We are entering a phase where the capabilities of AI are advancing faster than enterprises can absorb them.” “An AI agent that answers a customer query is relatively straightforward. An agent authorised to approve refunds, modify financial records or execute business transactions is an entirely different engineering challenge,” he notes. The main challenge is no longer how intelligent AI models are, but whether enterprises have reliable data, effective integration with existing systems, clear decision-making rules, dependable operations and a sustainable cost structure to support their deployment. “From where I sit, the next phase of enterprise AI will be defined less by how intelligent our models become and more by how much consequential business work we can reliably delegate to them,” says Gupta. “The real slowdown isn't necessarily in AI’s ability to act. It's in our ability to trust, govern and operationalise those actions at scale.” The distinction matters as businesses move from using AI to assist employees to letting AI agents execute tasks independently. An agent generating a response requires a different level of control from one authorised to modify financial records or approve transactions. For enterprises, this transition depends on reliable integrations, clearly defined permissions and the ability to manage the financial and operational risks of autonomous systems. Frequent model releases can delay implementation Jyotishko Biswas, an AI leader at HP Finance in India, points to another challenge: continuously reassessing AI models as newer versions become available. He acknowledges that newer AI models can offer benefits and should be evaluated, but once an end-to-end solution is finalised and implementation begins, frequent reassessments can disrupt progress. “There is continuous reevaluation due to new models coming, which slows implementation, and a lack of clarity for the execution team,” he says. Echoing similar views, Gupta says that organisations should not adopt newer models simply because they are available. Gupta notes that his team considers switching to a newer AI model only when it offers cost savings, lower latency, better response quality or when support for the existing model is discontinued. “We are trying to avoid industry pressure by adopting a new model instantly unless there is a need for a higher version.” Business value takes priority Taken together, the responses suggest enterprise AI is entering a phase where deployment and business outcomes matter as much as model capabilities. Organisations must determine where AI can deliver measurable returns, establish governance frameworks, prepare their data and systems, and decide when upgrading to a newer model is justified. The focus is also shifting from AI-assisted workflows to systems that can execute business tasks independently. That transition requires greater confidence in reliability, security and accountability. For enterprises, the question is increasingly not how quickly they can adopt the latest AI model, but whether they can put it to work reliably, responsibly and at a sustainable cost.
document fetched 2026-10-30 02:57