AI agents are starting to run real parts of a business, from finding customers to processing complaints. For banks, fintech companies, and enterprises, that raises a basic question: how far can an agent be trusted to act on its own?

Pine Labs is addressing this question by building an ecosystem that enables AI agents to take on tasks across commerce while keeping their actions accountable. Speaking at its product showcase at TechSparks 2026, Sanjeev Kumar, CTO, Pine Labs, said, “We have tried to make AI a first-class citizen in our ecosystem.”

Over the past 18 months, Pine Labs has been building a commerce operating system around AI agents, combining tools to manage their permissions and track their actions with applications for merchants and enterprise teams. Its Agent Store will open the ecosystem to third-party developers building for merchants and banks.

For agents to operate in banking and enterprise environments, organizations need to know what they are authorized to do and how their actions can be audited. Pine Labs has developed Grantex, which Kumar described as ‘Open Authorization 2.0 for AI agents’. It assigns each agent a defined set of permissions, an expiry, and a record of the human who authorized it. If an agent authorized to read and draft emails attempts to send one, the action can be blocked and its permissions revoked, including those of any child agents. A tamper-proof, hash-chain audit trail records its actions.

Kumar outlined the work involved in connecting tools and systems, setting up retrieval-augmented generation (RAG), running continuous evaluations, and deciding where human intervention is necessary.

Source link