Databricks: How Manufacturing AI Agents can Make Decisions

For many industrial businesses, AI remains an aspiration rather than an operational reality.

Shiv Trisal, Global Industrials GTM Lead at Databricks, asks: "How do we go from AI strategy being on a PowerPoint to being actually used in production, making real decisions alongside humans?"

That question is the lens through which Databricks, the data and AI company, frames everything it builds.

Founded in 2013, Databricks now serves more than 20,000 customers, including 70% of the Fortune 500. Its platform unifies structured and unstructured data under a single environment, then layers intelligence on top so that ordinary business users, not just data scientists, can act on it.

Despite its scale, Shiv says a founding instinct has not left the business: "We still operate with the mentality that we were born yesterday, even though we are at 10,000 to 12,000 people now."

Read Shiv's full interview.

Data fragmentation

Before any AI model can deliver meaningful results, the data feeding it must be clean, connected and accessible. That is easier said than done in manufacturing and supply chain, where data is spread across dozens of disconnected systems.

Mal Vivek, Founder and Chief Executive Officer of AI consultancy Zeb, has seen this problem at first hand: "It's across so many different systems and it's very heterogeneous, structured, unstructured, semi-structured in some cases."

The consequences are significant. Without a cohesive data foundation, proof-of-concept AI projects stall, leadership loses faith in the technology and nothing reaches production.

Anupam Gupta, Co-Founder of Celebal Technologies, says: "Some of my manufacturing customers have as many as 44 ERP instances, with around 33 unique software tools." His team builds on Databricks to consolidate those environments and deploy AI agents capable of taking corrective action across them.

Breaking down walls between data

To solve the problem at a structural level, Databricks is advancing a new architecture it calls Lakebase. For decades, analytical data systems and operational systems have lived apart, forcing businesses to manage costly, error-prone data extracts between them.

Shiv explains: "That world of taking extracts from the analytical world into the operational world accepts that humans will be doing work on spreadsheets. Agents, unfortunately, don't work like humans."

Lakebase dissolves that boundary, allowing AI outputs to flow directly into live operational workflows without manual intervention. The automotive sector illustrates the opportunity well. Managing a bill of materials across tens of thousands of parts and multiple vehicle models, while reconciling live risk and recall data, is precisely the kind of problem Lakebase is built for.

Shiv draws on his experience from aerospace: when his team could predict a likely aircraft component failure, the next steps of checking parts availability and scheduling engineers still required querying entirely separate systems. Lakebase changes that.

Bringing data intelligence to every user

Databricks' most visible product for business users is Genie, an agentic AI system that answers questions about enterprise data in plain language.

A user at an energy company can ask why container vessels carrying a particular product are off-route. Genie builds a multi-step analytical plan in the background, maps it to curated company data and returns an answer without the user writing a single line of code.

Shiv uses it daily himself: "I'm responsible for 3,000 customers at Databricks and I've replaced all my Excel sheets and 40 dashboards into one conversational interface."

The system also learns the language of each business over time, incorporating company-specific terminology through usage and human feedback. For a manufacturing executive who needs to understand operational risk, not sensor-level noise, that kind of contextual intelligence matters enormously.

Partners are closing the last mile

A platform is only as powerful as the ecosystem built on top of it. Several Databricks partners are now applying its capabilities to some of the most persistent problems in manufacturing and supply chain.

Rakesh Sancheti, Chief Growth Officer at Tredence, describes his company's focus as solving the "last mile" problem in AI, the gap between generating an insight and embedding it into a business workflow. Tredence's RAPID platform uses multi-agent systems to cut decision latency and eliminate manual steps across departments.

Kinaxis, the supply chain planning software company, is integrating Databricks directly into its Maestro platform.

Andrew Bell, Chief Product Officer at Kinaxis, explains that agentic AI is driving three outcomes: autonomous decision-making, democratisation of access for non-technical users and composability of end-to-end workflows. Shiv explains: "It's not just humans monitoring dashboards, it's humans and agents collaborating to achieve outcomes."

The physical world

Shiv is candid that AI has so far fallen short of its promise in industrial settings: "Frontier models haven't really made much of an impact in the physical world. It starts with connecting machines, data and people and embracing that in how you get work done."

The priorities for Databricks over the coming year are clear: rolling out Genie more broadly to business users and establishing lakebase as the foundation for human-agent collaboration. Both depend on bringing IT, operational technology and engineering data together under a single intelligence engine.

The question is not which model to use, as everyone has access to the same frontier models now. The question is what data a business holds and how well it can put that data to work.

That, in the end, is where competitive advantage is built.

Read the full interview.