At COLLIDE 2025, IBM's Ryan Zombo laid out five questions enterprises must answer to move agentic AI into production, from avoiding agent sprawl to avoiding lock-in. He argued that governance works like brakes on a race car and shared productivity results from IBM's internal client zero program.
Key takeaways
- Five questions come first. Agent sprawl, unstructured data, governance, hybrid deployment and avoiding lock-in.
- Data is the fuel. Zombo said only about 1% of enterprise data is used by generative AI today.
- Unify the data platform. Access to all data, multiple integration patterns, a lakehouse, lineage and life-cycle security.
- Governance is the brakes that let you go fast. Zombo called for one centralized approach across tools and stakeholders.
- IBM tests on itself first. Its client zero program reported $3.5 billion in productivity gains since 2022, he said.
Moving enterprise AI from experiments to production comes down to three layers working together: agents, trusted data and governance, according to Ryan Zombo, Director of Data and AI Technical Sales at IBM. In a keynote presented by IBM at COLLIDE Data + AI Conference 2025 in Atlanta, Zombo, who said he leads IBM's data and AI technical team for the Americas, laid out the questions enterprises must answer and shared results from IBM's internal AI program.
What questions must enterprises answer to scale AI?
Zombo traced the path from the release of ChatGPT around 2022, through rapid experimentation, prompt engineering and retrieval augmented generation, to the arrival of agents that can plan, decide and act. Taking that to production, he said, means answering five questions:
- "How do we scale with agents across various verticals without agent sprawl?"
- How can the business use unstructured data from every source, not just traditional data?
- How will it govern the data, the models on top of it, the applications that use them and the agents that take action?
- How can it do all of that wherever it needs to, in its own data centers, on hyperscalers or across clouds?
- How can it keep pace with change? Models and providers come and go, he said, and companies want to adopt new innovations without being locked into one technology.
IBM's answer, Zombo said, is a unified platform for building and managing agents, built on trusted data and strong governance.
What makes an agentic AI platform enterprise-ready?
Agents depend on the tools that connect them to company systems and third parties, Zombo said. A solid platform starts with a catalog of pre-built agents that teams can tailor to their business. Those agents must be high quality and dependable, able to scale from tens to hundreds to thousands, and secure enough to put in front of employees and customers.
He framed the shift through conversational AI's history. Early chatbots were robotic and pre-programmed, and customers pressed zero to reach a person as fast as they could. With generative AI, he said, "self-service is a good experience," and agents are the next step.
Why is enterprise data the fuel for AI agents?
Zombo compared agents to a Formula 1 car that goes nowhere without fuel. Data is that fuel, he said, yet "only about 1% of enterprise data is currently being leveraged by generative AI use cases." Most enterprise data is unstructured, and running on models trained on data available to everyone won't differentiate a business.
He listed the pitfalls that stall AI projects: low accuracy caused by unlabeled data and a poor understanding of relationships, missing lineage and data quality, inconsistent governance, and weak security and access controls over sensitive data.
To address them, Zombo described the capabilities of a unified data platform:
- Access to structured, semi-structured and unstructured data
- Multiple integration patterns, including batch, streaming, change data capture and replication, with visibility into pipelines
- A lakehouse that combines data lake flexibility with data warehouse reliability, with fit-for-purpose engines and open table formats
- Data intelligence on quality, origin and lineage, so data can be shared as a product with BI and AI applications
- Security across the life cycle from ingestion to consumption, running on hybrid cloud
Why does AI governance help teams move faster?
A colleague once asked him why cars have brakes, Zombo said. The answer: "You put brakes on a car so that you can go fast." Governance works the same way for enterprise AI, he argued.
Teams face changing regulations, a mix of homegrown, open-source and vendor tools, and many stakeholders. To move quickly, he said, "you need a unified and centralized approach to governance," with central visibility into how models perform, how they were built and what data went into them. He pointed attendees to a public case study on IBM.com describing how IBM governs AI internally.
What results has IBM's client zero program produced?
IBM's internal AI program, which it calls client zero, applies its own AI and automation across lines of business with the goal of becoming one of the most productive companies in the world, Zombo said. Coming into 2025, IBM announced $3.5 billion in productivity gains since starting the initiative in 2022, and he said it projected $4.5 billion by the end of the year.
The approach is to simplify end-to-end workloads, automate wherever possible and embed AI everywhere, Zombo said. Savings fund further investment, and running the technology at IBM's scale stress-tests it before customers adopt it. Examples included summarizing software support cases, generating marketing content for more personas and generating code for playbook development.
The internal AskHR assistant handles 94% of employee requests through AI before a person needs to help, he said, at lower cost and with a 55-point improvement in net promoter score.
Zombo said IBM may not be the most productive company in the world yet, but is making progress, and urged attendees to look for similar opportunities in their own lines of business.
Hear from enterprise AI leaders at COLLIDE 2027
COLLIDE, the largest AI conference in the Southern U.S., brings together more than 1,000 enterprise data and AI leaders. COLLIDE 2027 takes place Wednesday, October 13, 2027, at the Sandy Springs Performing Arts Center in Atlanta. Registration is coming soon.
Frequently asked questions
How do enterprises scale AI agents without agent sprawl?
Zombo recommended a unified platform with a catalog of pre-built agents that teams tailor, built on trusted data and centralized governance, able to scale from tens to thousands of agents.
How much enterprise data is used by generative AI?
Zombo said only about 1% of enterprise data is currently leveraged by generative AI use cases, and most enterprise data is unstructured.
Does AI governance slow innovation?
Zombo argued the opposite: like brakes on a car, a unified and centralized governance approach lets enterprises adopt AI applications faster.
What is IBM client zero?
IBM's program for applying its own AI and automation across its business; Zombo said IBM announced $3.5 billion in productivity gains since 2022 coming into 2025.
Watch: Propelling the Next Wave of Enterprise AI, presented by IBM on YouTube
Event: COLLIDE AI Conference · More coverage: COLLIDE AI Conference insights
