Salesforce Says Indian Companies Are Taking AI Agents Into Production
At Dreamforce 2026, Salesforce's South Asia leadership said Indian firms have moved past AI pilots and into production, with financial services leading. The harder part now is governing the agents.
For the last two years, most enterprise AI conversations in India have ended with the same word: pilot. According to Salesforce's South Asia leadership, that is starting to change.
Speaking to PTI on the sidelines of Dreamforce 2026 in San Francisco (15–17 September), Arundhati Bhattacharya, President and CEO of Salesforce South Asia, said Indian companies are now putting AI into production rather than just experimenting with it. She described India as having a lot of "white space" for digital and agentic AI, and said Salesforce will keep investing in infrastructure, hiring and upskilling here. A new Salesforce tower in Bengaluru is expected to open next calendar year.
Financial services is going first
Arun Parameswaran, who runs sales and distribution for South Asia, said the move from experiment to production is fastest in financial services. His examples were specific: vehicle-loan sales and credit underwriting, where work that used to take hours or days is now done in minutes.
He also described the order in which companies are deploying agents. Lower-risk tasks come first, the ones where a mistake doesn't cost much. More consequential decisions follow as trust, security and the agent's understanding of business context improve. He called this the normal maturity cycle, and it matches what I see in large Indian enterprises. Nobody starts by handing an agent the credit decision. They start with the paperwork around it.
The real work is managing the agents
Rohan Kumar, Salesforce's President and Chief Platform and Engineering Officer, made the point I think matters most. If an enterprise ends up with a million agents, someone has to manage them. How do you know they aren't going rogue? They need identities, monitoring, cost controls and a data layer that carries business context.
His view is that IT leaders will end up owning the number, activity and cost of agents, while security teams will issue agent identities and watch what the agents do. He said this is already happening.
Salesforce frames its own answer as a four-layer "agentic enterprise" stack: Data 360, Customer 360, Agentforce and AIforce. Bhattacharya added that the company will keep buying capabilities it doesn't build, pointing to its acquisitions of customer-agent company Fin and data management firm Informatica, and wrapping them in its trust layer.
What changes for the workforce
The executives were direct about jobs. Bhattacharya expects AI to take over repetitive work and push people toward more specialised roles, naming AI ethicists, prompt engineers and forward-deployed engineers as examples. Parameswaran put it more bluntly: work that once needed 20 developers might need two, done in a tenth of the time, even as other roles get created.
Kumar ended on India's developer base. Looking at where the most advanced agentic-enterprise work will happen by 2030, he said he believes India will be at the top of the list.
My take
Vendor executives at their own conference will always sound optimistic, so I'd read the "top of the list" line as ambition rather than a forecast. The more useful signal is the sequencing Parameswaran described, and Kumar's warning about governance. Moving to production is the easy headline. Knowing how many agents you have, what each one can touch and what it costs you is the part most Indian enterprises haven't built yet.