IBM wants to look beyond ‘just’ AI. The company has set out to merge existing workflows and automation with the management of AI agents, regardless of whether or not they’re built on IBM’s own platform.
During IBM’s Put AI To Work event in Utrecht, we spoke with Parul Mishra, IBM Vice President of Global Sales, among others. Although we will delve deeper into IBM’s day-to-day work with customers later in this article, we will first discuss the landscape of agentic AI and the role the company wants to claim.
Orchestration layer
Mishra starts from a familiar but nevertheless important point. AI-driven interactions in natural language enable employees without a high level of expertise to work within all kinds of systems. If the relevant solutions communicate well with each other, an employee could, for example, change their own phone number without having to go through multiple screens and system commands. These could be brand-new AI agents that simplify otherwise complex tasks, going beyond process automation by flexibly handling end-user requests. We will come back to this later.
First, Mishra points out a fact that organizations need to take into account more and more. Tomorrow’s employees will have grown up with ChatGPT and the natural language interaction enabled by the underlying GenAI technology. They will expect this simplicity in their working lives. That is why, according to Mishra, just about every IT vendor is developing a built-in AI assistant. “But now every application has its assistants and agents. The next step is: I host a marketplace, or an aggregator for all agents.” An example is Salesforce with Agentforce, but ServiceNow and Atlassian are also making moves in this area. Mishra notes that they all want to be the central platform on which all other AI agents come together, or on which all AI agents are built. IBM sees an opportunity to bring order to this impending chaos.
This is reflected in an existing solution that has been worked on extensively over the past few years: IBM watsonx Orchestrate. The underlying Orchestration Engine is designed to capture complex processes in a single experience. Within this, processes can be simplified through automation, sometimes with AI agents in the mix, but sometimes without. Based on first-hand experience, IBM also has constructed pre-built assets within this Orchestrate platform. It started utilizing a more granular automation with its own HR department. “What questions did the approximately 300,000 IBM employees ask most frequently over the past three years?” was the starting point. This resulted in 80 use cases for which assets are now available, says Mishra. One example: MFA can now be set up without the intervention of another person thanks to an agent. IBM will add more assets for other areas, but HR was an obvious starting point.
“Pre-built does not mean ready-made,” Mishra points out. “You always have to tweak it for your own business environment, but at least you don’t have to start from scratch.” She expects organizations to experience a “long tail” after adopting pre-built assets. They will increasingly place more tasks in Orchestrate.
Others are shouting louder
This sounds familiar. Companies such as Salesforce, ServiceNow, and SAP seem to be shouting about such functionality from the rooftops a lot louder than IBM. Mishra warns that they are all thinking from their specialization, be it CRM, ITSM, or ERP. That ball bounces back just as hard (“Isn’t IBM the mainframe specialist?”). However, she immediately adds another important argument that strengthens IBM’s position. “Everyone starts from a core set of principles and differentiation.” As an expert in business processes and an experienced player, the company has a rock-solid set of partnerships that its competitors sometimes lack. What’s more, IBM can glue together missing integrations. The work of linking APIs is a hellish task, but Mishra seems to suggest that IBM is better equipped to do this than the rest of the market.
The strategy is somewhat reminiscent of AWS’s Bedrock promises from late last year, but that is less interesting for a hybrid cloud approach. Mishra sees that the hyperscalers usually supply the building materials, while IBM intends to have the construction work already done. As for competitors in other areas, Mishra says that other companies’ AI agents are also welcome on watsonx Orchestrate. “We don’t want to be Salesforce. Their strength lies in managing sales processes and data.”
It is to be expected that AI will primarily be used to accelerate existing work processes in the coming years. This makes it obvious that you will apply it within existing IT solutions. Today’s term is “agentic,” where automation goes a step further than before because AI tooling can take context into account and make dynamic decisions. Mishra acknowledges this, but points out that deterministic workflows will always exist and be necessary. These are not based on probability calculations like GenAI. “You don’t want AI to carry out your promotion process or handle your invoice processing,” she gives as an example. In other words, such matters will remain hardcoded to provide certainty. Of course, AI can assist in creating these processes, but it should not have to reinvent the wheel every day.
According to Mishra, what an AI agent does must also be transparent to the end user, for example via an LLM chatbot. She gives an example of a user who wanted to change their phone number in the IT systems. The agent found the location where this number needed to be changed, but came across a field with a default number in it. The user had not mentioned this, as the question was far too general. (Moreover, did the employee in question even know that this field existed?) Ultimately, the agent chose to enter the new phone number in this field. Mishra: “The interesting question is: did you want this to happen? In this case, the decision is not too important. But wouldn’t you want your agent to ask the user if they explicitly want this to be done?”
Therefore, The answer to how autonomous your agent should be will depend on the use case. As Mishra explains, it is a trade-off between control and flexibility, whereby you may take steps towards autonomy as you gain more trust in the agent. But always with a human being at the controls to keep things in check.
Policy issues
Finally, Mishra discusses the governance layer—the “policy engine” used in other suites, such as those from Salesforce and ServiceNow, to ensure that AI agents remain compliant and work with the right data. IBM does not provide a separate product in this area, but weaves the control layer directly into watsonx Orchestrate. Within the platform, these security controls are called patterns. They started out as simple routing rules, but now also include reasoning and governance logic. Eventually, the same layer will even suggest the next process steps; IBM refers to this as automation planning.
These patterns are based on familiar foundations such as workflows, content management, intelligent document processing, and process mining – “boring stuff,” says Mishra jokingly, “but that’s exactly what businesses run on.” By adding such building blocks to Orchestrate, an agent can make informed decisions and sometimes even decide that no agent is needed at all and that a classic automation workflow is sufficient. “We don’t just orchestrate agents and assistants, we also orchestrate the workflows and decision moments that surround them.”
Orchestrate is now generally available. IBM is emphatically not positioning it as a marketplace, but as a mature layer on which organizations can securely enable their existing and future AI agents to work together.
From promise to practice with Client Engineering
To make the leap from technology promise to concrete business value, IBM relies heavily on its Client Engineering teams. In Utrecht, we speak with Peter den Haan, who has been head of the Dutch Client Engineering team for two years and has been with IBM for fifteen years, and Valentijn Stolk, who worked as a Technical Specialist AI Applications last year.
Den Haan succinctly summarizes their role: “We translate vague ambitions (‘we want to do something with AI’, ed.) into an answer that fits the customer’s real needs.” Each innovation process takes them no more than two months. The success stories that IBM presents during Put AI To Work almost always originate from such a process. A BI reporting request can be quickly defined; “something with AI” first requires a reality check of what problem a customer actually wants to solve. That is why IBM organizes workshops in which it works with the customer to identify where gains can be made in the processes. As soon as the discussion about costs begins, it simply revolves around what AI delivers in concrete terms: faster turnaround times, higher customer satisfaction, new sources of revenue.
This approach was successful last year at Damen Shipyards, as announced on stage at Put AI To Work. Their real pain point was quality assurance. Think of specifications that turn out to be incorrect after the fact. These errors cost a lot of money and delay construction, explain Den Haan and Stolk. Damen works with “mega documents” – enormous drawings that have to be translated into a Bill of Materials. Some parts have numbers, others don’t. IBM worked with Damen Shipyards to create a digital twin that interprets these documents and detects errors at an early stage. This equips a traditional manufacturing company founded in 1927 to deal with the problems of 2025.
Down to the smallest detail
Client Engineering delves deep into the organization to determine the scale of a process, embed new workflows, and apply AI where it makes sense. But, Den Haan warns: “You can’t replace your foundation every three months.” While the hype shifts from RAG to agentic to whatever hyped term will follow, bringing a single solid process into production yields more than starting over again and again. You can’t continuously incorporate every innovation in such a process. That’s especially true if you’re coming up with all kinds of custom solutions, Den Haan explains. That’s why IBM keeps the software side as standard as possible, with watsonx Orchestrate and watsonx in general as the backbone, so that innovations can be added without breaking down the whole thing.
Stolk compares Watson Explorer (and the current Orchestrate) to a framework in which a single task rarely resides in a single system. Employees constantly hop between apps; the platform brings those scattered steps together, now enhanced with agentic AI, RAG components, and decision logic. Crucially, the framework is expandable where necessary. For existing IBM customers, this is a logical next step after years of business automation. New, growing customers experience a quick kick-start, as Den Haan believes it is considerably easier to set up a completely new work process than to replace something.
Den Haan sees a pattern: growing companies experience the most growing pains and therefore have the greatest need for something new, while established companies struggle with, say, replacing five legacy systems. AI also serves as a magnet for young talent, he says. As Mishra already pointed out, the new generation of employees expects a different, intuitive way of working.
This brings us to the end of our discussion and shows us how IBM’s orchestration layer has been designed and set up in a practical setting. IBM positions watsonx Orchestrate as the layer on which the work processes of today and tomorrow must land. Although it is not a marketplace, we are told, it is ideally a collection point for all your efficiency gains. There are plenty of competitors who want to position themselves in the same way.