Inside NPCI's AtOM: Agentic Orchestration for UPI's Coordination Problem

The difficult part of UPI innovation is getting all parties to act in sync. NPCI’s AtOM (Agentic Orchestration & Messaging) is a framework built to solve exactly that: coordinating how changes to the UPI network move across banks, payment service providers and technology partners. The introduction of a new feature in the UPI is not finished once it is specified. After a change is specified, there is a lot of work to do. Financial institutions need to assess its implications for their own systems. Payment service providers need to understand the requirements and implement them. Technology partners have to test their solutions, reconcile discrepancies and prove their readiness for certification. All parties work towards the same goal, from different documents, timelines and understanding. This is the less visible part of payments infrastructure development. A change can be specified on the network level and still require continuous coordination to get implemented by all the parties that should implement it. AtOM is NPCI’s answer to this coordination problem.

From publishing a change to orchestrating it#

Traditionally, a network change triggers a chain of actions in the organizations involved. A specification is distributed, questions arise, clarification happens, implementations are made and certification is done. In the process, context is lost and dependencies remain unresolved. The harder challenge is keeping this work connected. AtOM treats this as a lifecycle of its own and not as a series of separate tasks. Its public platform covers the whole journey from research and requirements to technical specifications, partner implementation, testing and certification. AI agents assist at the individual stages, and evaluation checkpoints sit between them. The more important idea is how the participating organizations communicate. AtOM keeps the network platform, which creates and distributes changes, apart from the partner platforms, which receive the changes, assess their impact, ask questions and report readiness. The two sides communicate through a structured Agent-to-Agent (A2A) interface. So instead of depending only on people to turn documents into emails, meetings and status updates, agents can share structured information on requirements, queries, implementation progress and certification. Automating the documentation of one team saves effort in that team. Connecting work across organizations is a different problem.

Interoperability without uniformity#

Payments infrastructure relies on participants with different systems, different technology stacks and different implementation constraints. It is neither realistic nor needed to ask all of them to work in the same way. What counts is a common way to share information and show progress, while every organization decides how a change fits into its own environment. This separation is visible in AtOM’s public architecture. The network and partner platforms are deployed independently of each other and share one communication contract. Partners can adjust their implementation logic to their own systems without standardizing their whole technology stack. Here, agent-to-agent collaboration gives organizations a way to structure their cooperation without having to operate as one. The goal is to keep people in the process and reduce the effort spent on moving information between teams, so that decisions, dependencies and readiness are easier to track.

Trust is built into the workflow#

In payments, faster implementation has little value if control is lost along the way. Generating a specification or a code change does not prove that it is correct. An implementation still has to be reviewed, tested and certified. AtOM’s public design has evaluation gates and keeps code integration under human control. Its documented build workflow does not deploy generated code automatically. This is an important principle for enterprise AI. Autonomy should be limited by clear responsibilities and verifiable results, especially when institutions depend on each other’s work.

The infrastructure behind the innovation#

UPI’s evolution is described in terms of what customers can do: make a payment, set up a mandate or use a new feature. Every new capability depends on the work needed to bring the whole ecosystem along. That work deserves attention as well. AtOM’s value is that it treats ecosystem change as something that can be structured, coordinated and evidenced across organizational boundaries. Its open-source release also gives the technology community a reference implementation to study and build on. The public edition may differ from NPCI’s internal deployment. The next phase of payments innovation will depend on how reliably the ecosystem adopts change together, and not only on what single institutions can build. The real test of infrastructure is whether the whole network can move forward with what it enables.