In a move that signals the transition from “software” to “active intelligence” in fintech, Razorpay officially launched Agent Studio at its flagship FTX 2026 event on March 12, 2026. Built using the Claude Agent SDK from Anthropic, the platform allows businesses to deploy autonomous AI agents that don’t just “talk” but “act” directly on top of payment and banking infrastructure.
Intelligence on Demand: The “Workforce” of Agents
The Agent Studio is designed as a B2B marketplace and a no-code builder platform. It enables merchants to deploy specialized agents to solve high-friction operational bottlenecks that previously required manual teams.
The “Initial 8” Production-Ready Agents:
- Abandoned Cart Conversion Agent: Uses voice-led interactions (via Zomato’s Nugget or SuperU) to call customers who drop off at checkout and offer real-time incentives.
- Dispute Responder Agent: Automatically monitors chargebacks, gathers evidence (delivery logs, activity signals), and submits optimized responses to increase win rates.
- Subscription Recovery Agent: Uses ElevenLabs voice synthesis to call customers about failed recurring payments (due to card expiry or low balance) and resolve the issue mid-call.
- Cashflow Forecaster Agent: Predicts liquidity gaps 3–7 days ahead, alerting finance teams to potential payroll or payout risks.
- RTO Shield Agent: Uses LLM-based address validation to detect high-risk “Return to Origin” orders before dispatch.
Beyond the Dashboard: The Agentic Experience Platform
Alongside the Studio, Razorpay introduced the Agentic Experience Platform, an AI-native layer that replaces traditional static dashboards with a conversational interface.
- Agentic Onboarding: Reduces merchant setup from 30+ minutes to under 4 minutes by automatically verifying identities against government infrastructure via a simple conversation.
- Agentic Integration: Allows developers to connect Razorpay to their tech stack in under 10 minutes using AI coding environments like Claude Code and Replit.
- Conversational Dashboard: Merchants can manage reconciliation by simply uploading a bank statement screenshot and asking the AI to “match these with my settlements.”
Technical Edge: The “Model Context Protocol” (MCP)
The platform’s autonomy is powered by Razorpay’s MCP Server, which acts as a “translator” between LLMs and payment rails.
- Zero-Code Builder: Businesses can create custom agents by describing tasks in plain English (e.g., “Follow up on unpaid invoices over ₹10k every Monday via WhatsApp”).
- Deep Integration: Agents natively connect with Shopify, WhatsApp, Shiprocket, Tally, and QuickBooks, providing them with full operational context to make smarter decisions.
A New Operating Layer for Business
“Businesses don’t just need more software anymore—they need intelligence that can act,” said Harshil Mathur, CEO of Razorpay. “With Agent Studio, a single person can operate like a team of 100 agents.”
The launch has already seen major pilots with Indian giants like Zomato, Swiggy, PVR INOX, and bigbasket, who are using “Agentic Payments” to enable customers to discover, decide, and pay entirely within a single conversational or voice-led interface.
| Metric | Legacy Platform | Razorpay Agentic (2026) |
| Onboarding Time | 30–45 Minutes | ~3 Minutes |
| Integration Time | 5+ Hours | <10 Minutes |
| Manual Operations | High (Human-led) | Low (Agent-led) |
| Interface | Clicks/Navigation | Natural Language/Voice |
