The episode opens with Apurva welcoming listeners to a MongoDB Podcast episode focused on powering agentic AI with Capgemini and MongoDB. Vinay Makkaji introduces himself as part of Capgemini's Cloud Center of Excellence, with 23 years in IT, Fortune 500 solutioning experience, and almost 5 years working with MongoDB. Farid Mohammad describes his 3 years at MongoDB supporting partner ecosystems across the Americas, working with tech services firms, hyperscalers, and ISVs to build MongoDB Centers of Excellence and reference architectures. Together, the introductions establish the episode's core dynamic: deep practitioner experience on both the SI and platform sides of enterprise AI transformation.

Apurva opens the substantive discussion by asking where GenAI is actually moving the needle — and what the biggest gaps remain. Vinay sets the scene: years of prototype-building with AWS, Google, and Microsoft are finally translating into real production deployments, with frontier models from OpenAI, Anthropic, and Google DeepMind now pushing toward reasoning and multimodal understanding. Farid then maps the journey into three crisp phases. Phase one (2023) was experimentation — connected chatbots and Copilot pilots, impressive but disconnected from enterprise systems. Phase two (2024) was enterprise integration — corporations realizing that operationalizing AI required confronting the messy reality of fragmented legacy data, which is where RAG architecture and knowledge graphs came to the fore. Phase three (now) is agentic orchestration: AI that doesn't just answer questions but autonomously executes tasks and drives business outcomes. ^{[1] — Farid Mohammad "Enterprise AI has moved through three distinct phases. Experimentation in 2023 produced disconnected chatbot pilots. 2024 brought enterpris…" 05:50 Play} The sharpest insight comes from Farid's framing: the challenge was never the model itself, but the business-proprietary data needed to augment it and reduce hallucinations. ^{[2] — Farid Mohammad "When companies tried to move AI from pilot to production, they discovered the model wasn't the hard part. The real challenge was integratin…" 07:00 Play} Vinay adds that AI has evolved from merely responding to prompts to enabling auto-healing systems — shifting from proactive to predictive operational modes.

Apurva pushes into what a truly effective technology partnership looks like in the GenAI era, and both guests make the same core point: AI must be thought of as a system, not a model. ^{[1] — Vinay Makkaji "No single vendor can build a complete enterprise AI system. The most effective architectures combine LLMs, RAG pipelines, hyperscaler servi…" 10:20 Play} Vinay uses the example of a connected fleet solution built jointly with Google and MongoDB to illustrate how what used to require ten separate enterprise integrations can now be consolidated into a unified solution — though it still requires multiple specialized partners. Five components are named explicitly: the LLM, the RAG architecture, the trained models, hyperscaler-native services, and MongoDB's vectorization layer. Farid extends this by pointing out what MongoDB alone cannot supply: industry-specific domain knowledge, transformation governance, and the ability to span people and processes across a large organization. That is exactly what Capgemini brings. ^{[2] — Farid Mohammad "The companies that succeed in this AI transformation, they consider this AI as a system rather than just a model." 13:30 Play} The joint value proposition, Farid argues, is enabling customers to move from experimentation to enterprise-scale AI faster while reducing risk. Vinay closes with a philosophical observation: the future isn't AI replacing IT, but a joint venture between IT organizations and AI — co-evolving together.

The episode's centrepiece is a detailed walkthrough of a tri-party reference architecture built by MongoDB, Capgemini, and AWS for critical equipment predictive maintenance in the oil and gas industry. Farid sets the commercial stakes with a Baker Hughes study: 80% of oil and gas companies face at least one unexpected critical equipment failure per year, ^{[1] — Farid Mohammad "Unexpected equipment failure in oil and gas costs $250,000 per hour, and 80% of companies face at least one such failure per year according…" 17:10 Play} and when it happens, the cost is $250,000 per hour. The previous approach was entirely reactive — time-based, usage-based, or rule-based maintenance — with no ability to anticipate failure across a complex production pipeline involving third-party equipment from companies like Halliburton. The new architecture changes all of that. Sensors stream real-time telemetry to MongoDB. Historical failure records, maintenance notes, and semi-structured PDF equipment manuals are all vectorized and stored in MongoDB alongside live data. ^{[2] — Farid Mohammad "Sensors stream telemetry into MongoDB. Historical failure data and equipment manuals are vectorized and stored in the same platform. When a…" 20:30 Play} When an anomaly is detected in the telemetry, vector similarity search scans the historical failure corpus to find resonant past cases and surfaces a recommended repair approach. Finally, an AI agent automatically creates a service ticket and assigns it to a field engineer — proactively, before failure occurs. Vinay then describes how the solution matured: starting from basic RAG and workflow automation, it evolved to include digital twins, AI/ML hybrid models, IoT streaming, and ultimately the ability to predict faults even before anomalous data appears in the stream.

Apurva asks where the partnership is headed, and Vinay paints an ambitious multi-front roadmap. The first trend: customers are refusing to treat MongoDB as a separate transactional database. A Fortune 500 client migrating to a cloud lakehouse — whether on BigQuery, Azure Fabric, or Azure Data Warehouse — wants MongoDB embedded inside the lakehouse itself, enabling unified data management that spans streaming, object storage, and AI analytics in one estate. ^{[1] — Vinay Makkaji "A Fortune 500 client migrating to a cloud lakehouse isn't treating MongoDB as a separate transactional database — it wants MongoDB embedded…" 24:33 Play} The second major frontier is mainframe and AS400/DB400 modernization: decades of enterprise data locked in EBCDIC encoding can now be migrated to MongoDB using out-of-the-box tooling, unlocking that data for AI workloads. Third, digital manufacturing and industrial IoT — connected vehicles, fleet management, factory of the future — represent a 5-year growth runway beyond oil and gas. Farid adds the strategic dimension: MongoDB is no longer just a database choice for Capgemini; it is a co-investment partner protecting partner margins and accelerating multi-year transformation programs. In some cases, ^{[2] — Farid Mohammad "MongoDB is no longer just a database choice for Capgemini — it's a co-investment partner protecting margins and accelerating multi-year tra…" 28:26 Play} this model has already delivered 30 to 50% improvement in outcomes. The memorable closing line: technology builds capability, but partnerships build outcomes.

The final chapter brings both guests back for closing remarks. Vinay encourages listeners to reach out to the MongoDB–Capgemini team and frames AI not just as a productivity tool but as a large-scale organizational initiative reshaping how enterprises operate. Farid lands the episode's most quotable moment — and sharpest warning: AI initiatives will stall at the pilot stage if organizations haven't built a strong data strategy to support them. ^{[1] — Farid Mohammad "AI initiatives will stall at a pilot stage if you don't have a strong data strategy to support that." 30:18 Play} The alignment of data ecosystems, architectures, and people is not optional; it is the precondition for production AI. He closes with a phrase that crystallises the episode's thesis: 'technology builds the capability, whereas partnerships build the outcomes.' It is a fitting end to a conversation that consistently argued the model is the easy part — it is the surrounding ecosystem of data, domain knowledge, and partnership that determines whether enterprise AI actually delivers.