This article is sponsored by Arango and was written, edited, and printed in alignment with our Emerj sponsored content guidelines. Learn extra about our thought management and content material creation companies on our Emerj Media Services page.

​Scalable AI requires coherent, related context, since no mannequin can compensate for an enterprise structure that presents a number of variations of reality.

RAND Corporation found that greater than 80 % of AI tasks by no means attain manufacturing — twice the failure price of standard IT tasks — and recognized insufficient knowledge infrastructure and management misalignment as main causes.

Evidence of that problem spans industries and geographies. The Carnegie Endowment for International Peace noted in a January 2026 practitioner paper that many AI initiatives turn into trapped in “pilot purgatory” as a result of manufacturing environments require sturdy knowledge flows, governance frameworks, and institutional readiness that pilots not often take a look at. Solutions that reach one atmosphere typically fail to switch cleanly to a different, forcing organizations to rebuild important infrastructure repeatedly.

As brokers transfer from aiding workers to performing on behalf of the enterprise, governance turns into as vital as mannequin functionality. NIST’s AI Risk Management Framework stresses that reliable AI requires transparency, accountability, monitoring, and traceability all through deployment. Without these foundations, organizations can not persistently clarify, audit, or belief AI-generated choices.

The sample reveals that functionality is usually capped by structure fairly than intelligence.

In a latest collection on the AI in Business Podcast, Ravi Marwaha, Chief Operating Officer and Chief Technology Product Officer at Arango, and Sumedh Chaudhary, CTO US Industry Market at IBM, dug into why fragmented knowledge, lacking context, and brittle workflows push enterprises into an AI failure zone — and what it takes to construct agentic methods that act precisely and explainably inside actual, excessive‑stakes operations.

This article examines 4 insights that make clear why enterprise AI repeatedly stalls at scale —  and what leaders should rebuild to make agentic AI correct, explainable, and economically defensible.

• Unified context for correct agent choices: Agents want actual‑time context — what modified, how methods relate, and why it issues — to cease guessing and produce choices that match knowledgeable judgment.
• Fragmented Data the Root of the AI Failure Zone: Unified, choice‑prepared context constructed from related methods, fairly than consolidated copies, eliminates the architectural drift that drives AI failure.
• Regulated workflows as the proving floor for reliable AI: High‑stakes processes power the temporal consciousness, traceability, and proof requirements required to scale AI safely throughout the enterprise.
• Multi‑agent orchestration for finish‑to‑finish automation: Agents solely work once they can function over the identical contextual layer, enabling coordinated actions that substitute remoted pilots with actual workflows.

Listen to the full episodes:

Episode 1:  The Architecture Shift Behind Reliable Enterprise AI – with Ravi Marwaha of Arango​

​Episode 2:  How Unified Context Turns AI Into Real Enterprise Performance – with Ravi Marwaha of Arango

​

Guest: Ravi Marwaha, Chief Operating Officer & Chief Technology Product Officer at Arango​

Expertise: Enterprise AI, Product Strategy, Data Infrastructure, Digital Transformation

Brief Recognition: Ravi Marwaha is a expertise govt with greater than 30 years of expertise main product, engineering, and digital transformation initiatives throughout enterprise software program and monetary companies. He at present serves as Chief Product and Technology Officer at Arango, the place he leads product technique, engineering, buyer success, and implementation, with a give attention to AI knowledge infrastructure and enterprise AI platforms. Prior to Arango, Ravi was Chief Product Officer for the Commercial Bank Digital Platform at J.P. Morgan. He has additionally held senior management roles at Uptake, GE Digital, SAP, and Informatica, overseeing world product, engineering, cloud, and buyer success organizations. Ravi holds a Bachelor of Engineering in Mechanical Engineering from Karnatak University.

​Episode 3:  Why AI in Document-Heavy Workflows Fails Without the Right Foundation – with Sumedh Chaudhary of IBM

​

Guest: Sumedh Chaudhary, CTO US trade market at IBM

Expertise: Enterprise AI, Enterprise Architecture, AI Go-to-Market Strategy, Hybrid Cloud

Brief Recognition: Sumedh Chaudhary is a expertise govt and enterprise architect with greater than 20 years of expertise spanning AI, cloud, enterprise structure, and consulting. He at present serves as CTO for GSI & Industry Tech at IBM, the place he leads expertise technique and AI platform adoption by Global Systems Integrator partnerships. Prior to this position, he served as CTO for IBM’s GSI Ecosystem Partnerships and as Associate Partner for Strategic Deal Solutions. Earlier in his profession, Sumedh held consulting and management roles at Thomson Reuters and Deloitte, delivering enterprise expertise, knowledge, and analytics initiatives throughout the private and non-private sectors. His work has been acknowledged with a number of IBM honors, together with the IBM Architect of the Year Award for Watson. Sumedh holds each an MBA in Business Analysis & Marketing and an M.S. in Data Science from Georgia State University.

Unified Context for Accurate Agent Decisions

Ravi Marwaha opens the collection by arguing that manufacturing failures stem from the atmosphere round the mannequin fairly than the mannequin itself. Pilots function inside curated situations the place knowledge is pre‑chosen, workflows are simplified, and inconsistencies are shielded from the agent. Production introduces the full complexity of enterprise methods constructed for human navigation — CRMs, ERPs, ticketing instruments, knowledge lakes, and doc repositories — every holding a unique slice of reality.

“Your knowledge isn’t in a single place, and even organizations with mature lakes, warehouses, and MDM methods nonetheless function with structured, unstructured, and multi‑mannequin knowledge unfold throughout environments constructed for people fairly than brokers. When brokers are fed irrelevant info or stitched‑collectively fragments, they don’t turn into extra clever — they lose the grounding required to make dependable choices. Models hallucinate as a result of they don’t have context, not as a result of they’re weak.”

— Ravi Marwaha, Chief Operating Officer & Chief Technology Product Officer at Arango

Sumedh Chaudhary sees the identical dynamic in regulated workflows. In doc‑heavy environments, lacking context reveals up instantly as measurable error as a result of the agent can not keep continuity throughout pages or methods. Leaders typically misdiagnose this as an information‑high quality challenge, however each friends emphasize that the underlying failure is architectural: brokers can not make defensible choices when the operational info they depend on is incomplete or contradictory.

Executives seeking to operationalize this shift can start by clarifying what an agent should really know at the second of choice. In each conversations, three strikes emerged persistently:

• Define the choice context — establish the particular alerts an agent should depend on, not the whole historic report.
• Map the place that info lives — perceive which methods maintain which fragments of reality.
• Establish temporal consciousness — guarantee the agent can observe what modified and why it issues.

For leaders, the implication is evident: AI readiness will depend on whether or not an agent can entry the operational info required to grasp what modified, the place info lives, and the way methods relate. Without that basis, even robust fashions behave unpredictably as soon as they go away the managed situations of a pilot.

Fragmented Data the Root of the AI Failure Zone

Across each conversations, fragmentation emerges as a main barrier to dependable agent conduct at scale. Marwaha describes enterprises as environments formed by a long time of collected architectural choices — knowledge copied for comfort, methods layered atop methods, and reporting instruments constructed atop these layers. None of this was designed for autonomous choice‑making, and all of it turns into seen the second an agent tries to behave throughout methods.

“People mentioned they had been consolidating, however in follow they had been fragmenting much more — copying knowledge from one place to 3 locations, then ten locations, and finally consolidating these copies into an eleventh location. Over time, BI instruments created one more layer of knowledge, every model barely completely different from the final. This sample has been compounding for a really very long time, and brokers inherit all of these inconsistencies the second they attempt to act throughout methods.”

— Ravi Marwaha, Chief Operating Officer & Chief Technology Product Officer at Arango

Sumedh Chaudhary sees fragmentation manifest in a different way in doc‑heavy workflows. A mannequin can interpret a web page, but when web page‑to‑web page that means is misplaced throughout extraction or processing, error charges rise rapidly — particularly in regulated industries the place accuracy thresholds are express. The agent just isn’t failing to grasp the content material; it’s failing to grasp the relationship between fragments.

For leaders, the takeaway is that fragmentation just isn’t solved by centralizing every little thing. Centralization typically creates new copies and new inconsistencies. The actual problem is making certain choice consistency — tracing the place that means breaks between methods, variations, or paperwork, and treating these breakpoints as architectural points fairly than knowledge‑high quality issues.

Regulated Workflows as the Proving Ground for Trustworthy AI

Regulated, doc‑heavy workflows reveal enterprise readiness quicker than another atmosphere. According to Chaudhary, these workflows fail not as a result of the fashions are insufficient, however as a result of enterprises can not protect the continuity brokers have to cause throughout pages, paperwork, and methods. Error charges spike the second an agent loses the semantic thread that connects the workflow.

“A doc‑heavy workflow presents a 3rd degree of problem since you’re not coping with simply unstructured textual content — you’re coping with photos, tables, and web page breaks that disrupt the semantic thread. When methods flip from one web page to the subsequent, they typically lose the contextual layer that connects the info, and the agent can not reconstruct what the workflow requires. In regulated industries, error charges are measured explicitly, and instruments are deserted rapidly when these thresholds aren’t met.”

— Sumedh Chaudhary, CTO US Industry Market, IBM

These workflows power temporal consciousness, traceability, and proof into the structure. They require brokers to grasp what modified, when it modified, and the way that change impacts the choice at hand. When continuity breaks, compliance thresholds fail lengthy earlier than mannequin functionality turns into the limiting issue.

Chaudhary highlights a number of early indicators leaders ought to monitor earlier than trying scale:

• Error‑price enchancment week by week — stagnation alerts architectural points, not mannequin points.
• Temporal reasoning — brokers should exhibit consciousness of what modified and when.
• Semantic continuity — web page‑to‑web page and system‑to‑system that means should stay intact.

For executives, the implication is that regulated workflows should not area of interest — they’re the clearest diagnostic for figuring out whether or not an enterprise can assist agentic AI. If an agent can not keep continuity in these environments, it is not going to behave reliably anyplace else.

Multi‑Agent Orchestration for End‑to‑End Automation

The last perception throughout the conversations is that agentic AI turns into operational solely when a number of brokers can act over the identical related operational image. Marwaha notes that enterprises typically try and scale by including brokers, however with out shared operational info, every agent turns into one other silo. The problem just isn’t the variety of brokers; it’s the coherence of the atmosphere they function inside.

Sumedh describes how multi‑agent methods perform in doc‑heavy workflows. No single agent can keep context throughout extraction, web page transitions, doc relationships, and cross‑system that means. Instead, OCR brokers, vector brokers, splitter brokers, and matching brokers every carry a part of the reasoning load.

“In a properly‑designed system, you may have an OCR agent to extract the digital footprint, a vector agent to retailer embeddings, a splitter agent to protect web page‑to‑web page continuity, and an identical agent to hyperlink paperwork throughout the workflow. Each agent carries a unique a part of the reasoning load, and none of them can succeed alone as a result of the workflow will depend on how they work together. An excellent orchestration layer is sort of a live performance grasp — it coordinates the musicians, but it surely can not produce the efficiency with out them.”

— Sumedh Chaudhary, CTO US Industry Market, IBM

Orchestration binds these brokers right into a coherent chain of reasoning. Supervisor brokers can oversee choices, and separation of duties can scale back error in excessive‑stakes workflows. But orchestration solely works when all brokers function over the identical operational info — not separate pipelines or inconsistent copies.

For executives, the takeaway is that multi‑agent methods are an rising architectural strategy for coordinating complicated enterprise workflows. The issue just isn’t deploying brokers; it’s making certain they share the identical operational image of the enterprise.