From experimentation to execution: Industry leaders at Articul8 roundtable share what it really takes to scale GenAI

As the GenAI wave sweeps through the enterprise, the path from experimentation to scalable, sustained value remains elusive — especially in regulated, high-complexity sectors. To decode what it truly takes to operationalize GenAI, YourStory and Articul8 hosted a closed-door roundtable with senior technology leaders from India’s top enterprises. Participants from across automotive, manufacturing, digital infrastructure, and consumer technology, including Jubilant FoodWorks, JBM Group, Eureka Forbes, Wipro, Engineers India Ltd, VVDN Technologies, came together to unpack the realities of operationalizing GenAI in high-stakes environments.

The conversations – candid, contextual, and deeply practical – revealed the nuances of building expert-class GenAI systems that reason with enterprise logic. Moderated by Sangeeta Bavi, COO, YourStory, and anchored by Arun Karthi Subramaniyan, Founder & CEO, Articul8, the roundtable showcased firsthand lessons on everything from infrastructure and trust to use-case selection and cultural adoption..

1. From general-purpose to purpose-built: GenAI must speak the language of the enterprise

The panelists opened with a shared consensus: generic LLMs aren’t enough. For industries like automotive, infrastructure, retail, and manufacturing, GenAI needs to do more than answer prompts; it must reason like a domain expert. A recurring view was that GenAI, to be truly valuable, must be trained on context-rich proprietary data and understand internal systems, processes, and constraints.

Takeaway: Scaling GenAI in the enterprise begins with context. Domain specificity isn’t a feature but a prerequisite.

2. Embedding GenAI into workflows: Not another dashboard, but a co-pilot

POCs that sit outside day-to-day systems quickly become irrelevant. Real success stories emerged when GenAI solutions integrated into existing workflows, supporting operations like design reviews, shop-floor automation, supply chain intelligence, or after-sales support. The benchmark isn’t an impressive chatbot but a virtual co-pilot that understands service logs, reads compliance documents, and flags anomalies in real time.

Takeaway: GenAI earns its place when it becomes invisible, embedded in workflows and augmenting day-to-day decisions, not just delivering outputs.

3. Legacy systems aren’t going away; GenAI must work around and within them

Enterprises spoke openly about the realities of their infrastructure. From decades-old ERPs to fragmented engineering data, most environments are anything but clean. The ask from GenAI? Compatibility, not disruption. The most compelling use cases came from adapting GenAI to read structured documents, process legacy formats, and operate in restricted digital ecosystems.

Takeaway: GenAI success in complex enterprises hinges not on clean slates, but on the ability to adapt to legacy infrastructure.

4. Governance and trust: The non-negotiables for scale

Governance emerged as one of the most consistent threads across the conversation. Hallucinations may be the headline risk, but what truly keeps enterprises cautious is the lack of traceability, explainability, and robust version control. Leaders stressed the importance of knowing what data influenced an output, when it was last updated, and how the system performed over time.

Takeaway: Enterprises don’t just need GenAI to be smart; it needs it to be auditable, compliant, and traceable. Governance is no longer optional.

5. GenAI must deliver measurable business value

A pattern that emerged across sectors was the shift from experimenting with tools to chasing tangible business value. Enterprises have moved beyond vanity metrics like bug count or response time, and are now evaluating GenAI based on its ability to reduce downtime, improve throughput, or assist in faster decision-making.

Takeaway: Metrics like uptime and output don’t matter unless they lead to a business outcome. GenAI must tie into the bottom-line impact.

6. Adoption Requires Culture, Not just Compute

While compute infrastructure, training time, and model architecture were recurring technical topics, a more human issue also surfaced: adoption. Even in tech-forward organizations, internal resistance, especially from mid-level engineers and operations teams, can derail GenAI rollouts. Training, trust, and communication were highlighted as essential components for cultural alignment.

Takeaway: GenAI cannot succeed in isolation. It needs infrastructure, yes, but also people who trust and understand it.

7. Security and IP: Build first for isolation, then for intelligence

For many companies, especially those in engineering, design, or manufacturing, IP sensitivity is paramount. The roundtable underlined the importance of secure-by-design architecture, where no data leaves the system, customer context is ring-fenced, and models operate in air-gapped environments.

Takeaway: In high-stakes domains, data privacy isn’t a checkbox but it shapes the architecture itself.

8. Start Narrow, Scale with Proof

Leaders agreed that attempting to scale GenAI across the board rarely works. Instead, the winning approach is choosing one high-impact use case, validating outcomes, and then expanding. Whether it's predictive maintenance on the shop floor or document intelligence in legal ops, the roadmap must be intentional and iterative.

Takeaway: Start where the impact is undeniable. Prove value in one critical function before expanding GenAI’s footprint.

Articul8’s perspective: Moving from POC to production with purpose

Having spent years helping large enterprises bridge the GenAI gap, Articul8’s core proposition resonated clearly with the discussion in the room: most organizations don’t have a technology problem; they have a translation problem between business goals and AI implementation.

Articul8 deliberately avoids the word “POC,” choosing instead to run what it calls “production pilots” that are engagements that use real enterprise data, define clear business KPIs, and are designed to scale from Day 1. “We don’t get into a pilot unless you know what business value you’re chasing,” Arun Karthi Subramaniyan shared. This alignment-first approach ensures teams aren’t stuck in demo loops with no measurable outcome.

On the infrastructure front, he underlined that scalability is more about control and observability than compute. The platform supports air-gapped deployments, granular model versioning, and lifecycle auditability, making it suited for regulated, high-stakes environments like manufacturing, telecom, and public sector.

Data security and IP protection were also major concerns echoed by roundtable participants. Subramaniyan addressed this by clarifying that customer data stays local, and even logs aren’t visible to them. About 70% of their models are built from scratch, not fine-tuned off existing LLMs, giving enterprises greater confidence in compliance, data separation, and domain specificity.

Finally, on change management, Subramaniyan emphasized that the GenAI shift is as much about people as technology. The Articul8 team shared candid examples where engineers initially resisted GenAI adoption, assuming they could outperform the model, only to realize the potential of collaborative learning when paired with AI. The key, they noted, is structured experimentation backed by metrics that business stakeholders care about.

Final thought

The Articul8 India Enterprise Roundtable showed that GenAI adoption is shifting focus from tools to enterprise transformation. And that transformation isn’t hypothetical anymore. Leaders across industries echoed that the next chapter of AI isn’t about adding intelligence, but about embedding relevance.

That relevance is defined by systems that understand your plant, your codebase, your compliance landscape—not just the internet. It’s defined by use cases where GenAI drives root-cause analysis, flags regulatory anomalies, or accelerates product testing. And perhaps most importantly, it’s measured not in lines of code, but in downtime avoided, errors reduced, or decisions accelerated.