Share with your CIO
Enterprise AI is hitting an architectural ceiling, and the argument for breaking through it centers on orchestration rather than point-solution deployment. Writing for the Nasscom community, Tech Mahindra’s SAP Practice Head Avanish Kumar makes the case that most organizations are still running AI in silos, individual copilots, standalone agents, embedded predictions, none of it connected, and that the next competitive gap opens between companies that unify those signals and those that don’t. The post maps a three-stage path from embedded intelligence to autonomous multi-agent orchestration, with unified data models as the load-bearing layer throughout.
What this means for your business
The diagnostic here is more useful than the prescription. If your organization has deployed AI tools function by function, procurement has one, HR has another, supply chain has a third, you’ve likely already hit the ceiling the article describes. An agent that can’t see supplier risk data alongside contract terms and spend history will make worse sourcing decisions than a competent analyst with a spreadsheet. That’s not a futurist warning; it’s a current-state failure mode for any enterprise that bought AI capability before buying data coherence.
Kumar’s three-stage model, embedded intelligence to extended intelligence to autonomous orchestration, is a reasonable sequencing, but it quietly assumes the hardest problem is already solved by the time you get to stage three. Unified data models aren’t an implementation detail you stitch together on the way to multi-agent workflows. They’re a multi-year data governance effort that most enterprises are still in the middle of. The post, written by someone whose firm sells SAP transformation services, leans toward SAP’s Business Data Cloud and Joule as the connective tissue, which is a legitimate architectural path but not the only one, and the timeline optimism reflects that commercial angle more than engineering reality.
The CIO who should act on this isn’t the one asking whether orchestration matters; it’s the one deciding whether to bet on a single-vendor orchestration layer, SAP’s or anyone else’s, versus building interoperability across existing systems. Vendor lock-in at the data model level is a different risk profile than lock-in at the application layer. If the shared enterprise memory that makes multi-agent coordination work is owned by one platform, every future renegotiation happens on that vendor’s terms. That’s the renewal you should be repricing in your head right now.
Concept deep-dive: Multi-agent orchestration
Multi-agent orchestration is what happens when individual AI agents, each built to handle a specific task like contract review or demand forecasting, are coordinated by a governing layer that sequences their work, routes information between them, and enforces policy guardrails. Think of it as an air traffic control system for AI processes: no single plane decides the runway; the tower holds shared context about all of them simultaneously. Without that coordination layer, agents optimizing locally can conflict at the enterprise level.
Based on reporting from From Siloed AI to Orchestrated Intelligence: The Next Phase of Agentic Transformation | nasscom, originally published 2026-10-09 05:52:00.
