How IBM is Building the Future of Enterprise AI | Gartner D&A Summit Recap

AI isn't just an intern or an assistant anymore. It is able to do tasks independently and guess what that is the central theme in this year's Gartner's data and analytics summit of 2025. However, there's a very important point that a lot of the people overlook and that is it's not all about better models, it's about better data. IBM is a premium sponsor of this event by Gartner and their message is loud and clear. They have clearly saying that AI is no more an experiment. It's actually business critical. So let's go check out and talk to a few of them and also see what they're up to. The interesting part about agentic AI as it became uh more center stage people started to thought start thinking that if I have now agents taking decisions for me and they're doing work for me then my risk of execution is higher and hence I started to see conversations around governance now coming in saying that how will I govern an agent? How do you see Watsonx kind of improving the organizational productivity overall or AI agents for that matter? — Um the the perfect example for that is what we do in ICM itself. — Mhm. — A lot of the streams that we used to run whether it's HR, procurement, even sales. Um today are all run through agents. Now the reason that is very powerful is because earlier every seller who had to come in need to know how do they go in and create an opportunity who do they tag how do they progress through that opportunity a lot of times then what happens when a lot of their time was lost in going in and updating these systems — whether at the same time not being able to spend enough time with clients so people are asking oh they've been doing a lot of use cases on agent AI but he done pilots then after that goes in the bottom right it's just like that — so I think the key thing missing in this entire conversation is that there is no alignment with the business in this overall strategy of the organization so what happens is when every department as an from an enthusiasm point of view right okay agent a case today I want to try out something to show that okay something is working you try that and then you do right once that is done okay project is over something is presented then nobody knows what needs to be done with that So when you say governance right like uh please define that for me. You can put governance into the three buckets. — Okay. One is a pure play governing your inventory of models and uses. The second pillar of this is actually the risk management of your output. You put it right and interestingly if you see IBM actually came out with what we call risk atlas. — Okay. — The last one is a compliance. — Okay. That compliance is more around process compliance or a particular industry compliance. So we provide the tooling for you to really test the compliance of your model and your use cases to say that have they passed all the checks first of all as a process and then to assure automatically because it's a humongous task if you are bit on them at scale. — So today all that they're focused is having those conversations they come back and feed this information there is an RFP or a uh or a proposal that is automatically created. Now that's one of the uh problems that IBM sales team came back to us and said look while X does really great proposals Y is not able to do it. So we said no we give the same set of agents to everybody. So it's like having that one toolkit or a Swiss army knife which everybody has and that gives everybody the same set of tools and being product. What according to is the biggest uh agentic win IBM or anyone else has had which will make anyone watching this video feel that boss you have to double down on? Yeah, we have implemented for ourselves. — I know the HR, I know the sales, I know you know the procurement everything else, — right? So we done it for one of the largest pharmaceutical company and we it also in the L rings, right? Where they implement this technology for managing the regulatory uh reportings. — Okay, — because when they come out a new pharmaceutical formulas and all those kind of stuff. So they have to give lot of documentation to the regulator, right? uh what kind of research you have done, what are research results, say all the journals that you created, right? So much of documenting there to do it process takes 8 months time to complete start to finish. — These are the approvals of the new — approvals, right? And you have to talk to multiple departments, multiple agencies to get the data collected, validated. So here what we have done, we have created this agent solution which got skills in different domains. This can understand the particular let's say research areas coming to follow formulations right. So all those things form spot proive causes. Okay. Those skills are built and integrated. An agency system was implemented and now today we are able to do it in almost 2 months time. So from 8 month 8 months we reduce it to 2 months.

— 4x efficiency. — 4x efficiency — and I was having a conversation with client and he said that okay so explain me what will agent change for me right? — It's typically a 5minut exercise in order for you to build an agent and bring in your tools and start using it. The good part what I'm seeing with the gener the customers who have moved to production is that once you fine-tune your LLM right there is nothing like something which is saying that after a finetune LLM for a pilot I can't use it. The same LLM is going to be useful for you even for production also. So the pilots what you're doing or the PCs what you're doing are surely going to help you shorten your journey to production. One thing is very interesting is you know getting the stakeholders on the table from the very beginning. Who are your stakeholders right? When we talk about a project in a business, you know, the typical stakeholders are the business and technology, right? And operations as well because that's part of business. We have two previous booths where we talk about the data injection and how the data can be stored in a structured format the unstructured data. The second booth talk about uh how the agents can be built up. Here we are just trying to explain how the what what's the background around the agent what is behind the cover. — So uh as we have seen that AI has evolved rapidly right from you know machine learning to traditional AI now to agent AI right so the risk have been amplified as well. So when you talk about geni there is hallucination, drift, fairness, bias where you track these metrics. Similarly with agentic AI there are different tools LLMs prompts which you have to govern. This is typically used with a security landscape right with a lens of security. You literally have all these things listed down there right at the dashboard. You get to it on top of it and you're just there.