Companies should look beyond the efficiency gains from artificial intelligence and use frontier AI models to tackle problems they have historically been unable to solve, according to Irina Ghose, managing director, Anthropic India.
Speaking at the Global Fintech Fest (GFF) 2026, Ghose said companies were currently focused heavily on measuring how much efficiency AI could bring, while paying less attention to the more significant opportunities that could emerge from using the technology on difficult, open-ended problems.
“When you put these problems to frontier models, things that you don’t know, that’s where the intelligence shines,” Ghose said, citing drug discovery in healthcare and making sense of legacy code in banking as examples.
The argument comes as the cost of using increasingly capable AI models continues to fall, making it more feasible for companies to deploy them on complex tasks. Ghose said Anthropic’s latest model was about 25% cheaper for typical workloads and offered further savings for agentic, complex tasks.
Anthropic released Claude Fable 5.1 and Mythos 5.1 on September 1, with the company estimating that the newer models can reduce costs by up to about 45% for highly agentic workloads, driven by lower cache-read pricing.
Ghose also said companies needed to rethink the way they evaluate AI models. Traditional proof-of-concept cycles lasting several months could become counterproductive when new models are being released at a much faster pace.
“The fallacy of keeping a longer evaluation cycle is that by the time, that new model might have come out,” she said. “We released the last one about nine days back.”
Anthropic has released multiple model updates in recent months, underscoring the pace at which frontier AI capabilities are changing. Ghose said companies that were ahead of the curve were reducing lengthy pilot and milestone-review cycles and moving more quickly towards deployment.
The technology could also lower the experience barrier for specialised work, she said. Tasks such as understanding decades-old code in banks, which could earlier require months of work by senior specialists, can now potentially be undertaken in days with the help of frontier models.
For companies still hesitant about adopting AI, Ghose said the starting point should be leadership using the technology themselves and defining the business outcome they want, rather than beginning with a question about what AI can do.
However, access to the same frontier models means AI alone will not create a competitive moat. For listed companies, Ghose said differentiation would continue to come from their distribution networks, customer trust and data accumulated over time.