Artificial agents are beginning to perform real business functions, ranging from lead generation to complaint handling. For banks, fintech companies, and large enterprises, this raises a fundamental question: how much can one trust an autonomous agent?
Pine Labs is addressing this issue by creating an ecosystem that allows AI agents to perform commercial tasks while ensuring accountability for their actions. During a product presentation at TechSparks 2026, Sanjeev Kumar, CTO of Pine Labs, stated that the company aims to make AI a full participant in its ecosystem.
Over the past year and a half, Pine Labs has been developing an operating system for commerce around AI agents. This system integrates tools for managing permissions and tracking agent activities with applications for sellers and enterprise teams. Through its Agent Store, the ecosystem will be open to third-party developers creating solutions for banks and sellers.
For agents to function in banking and corporate environments, organizations must clearly understand the limits of their authority and have the ability to audit their actions. Pine Labs developed Grantex, which Kumar described as 'Open Authorization 2.0 for AI Agents.' This tool assigns each agent a specific set of permissions, an expiration date, and records the identity of the person who granted the permission.
If an agent authorized to read and compose emails attempts to send such an email, the action can be blocked, and its permissions revoked, including those of any subordinate agents. All actions are recorded in an immutable audit trail based on a hash chain.
Kumar detailed the work related to system and tool connectivity, setting up Retrieval-Augmented Generation (RAG), conducting continuous evaluations, and identifying moments requiring human intervention. These components must work in concert before an agent can reliably handle tasks in a corporate environment.
Pine Labs initially implemented this technology within its internal compliance, finance, HR, and marketing teams, running all operations in its own environment. The company is also implementing this technology in collaboration with banks in India and abroad.
The Cora agent, responsible for compliance and issue resolution at Pine Labs, handles cyber fraud complaints from receipt to full closure. It matches complaints with transactions, classifies them, routes them to relevant teams, and prepares responses for review. Kumar noted that this allowed Pine Labs to close some complaints the same day, whereas the standard timeframe is three to five days.
The agent logs a 15-step trail of its work, which includes data verification, transaction searching, reconciliation, team notifications, as well as supporting evidence and confidence levels. If the system connector is unavailable, Cora can utilize Robotic Process Automation (RPA). During a demonstration, it accessed the state cyber police portal and filled out an action report, specifying transaction details and the disputed amount. Meanwhile, a human operator maintains control over the password and one-time code and reviews the form before submission.
After deploying agents in internal compliance, finance, HR, and marketing departments, Pine Labs opened its platform to developers and seller partners. The company is also collaborating with banks in India and internationally.
To demonstrate the potential application of the technology beyond its own operations, Kumar mentioned a hackathon organized by Pine Labs in collaboration with The Ken. Over 380 participants—students, recent graduates, and engineers from India—took part. The winning project helped people decide what to cook for dinner.
Developers can create skill files, connect tools, and define tasks using plain language, while AgenticOrg provides transparency regarding agents, confidence scores, and audit trails via Grantex. AgenticOrg is designed to assist third-party developers in building seller and bank applications using Pine Labs' infrastructure.
For small businesses, Pine Labs has also developed an advertising agent accessible through the seller interface. Sellers connect their Google Ads account, provide information about their business, budget, and campaign goal. The agent researches local demand and nearby competitors, then generates ad copy for search, display, and YouTube.
Sellers can request changes and approve the campaign before launch. The demonstration showed that setup takes about two minutes, and campaign performance and lead routing are available through the same interface.
From internal enterprise workflows and cyber fraud resolution to small business advertising—Pine Labs applies its agent infrastructure to various aspects of commerce. Grantex establishes the framework for agent actions; AgenticOrg helps teams, individuals, and AI enthusiasts build, manage, and execute tasks with agents. Pine Labs is betting that the future of commerce will be determined not only by what AI agents can do, but by how reliably they do it. By combining agent management infrastructure with applications that solve real business problems, the company aims to make autonomous AI a working part of commerce for enterprises, banks, and sellers.
