I don’t want to interrupt, but
after some time working on AI agents, some factual facts I must share today
AI agents are not your employees and they are not to substitute people, but tasks. AI agents lack the adaptability of humans. They excel at following specific instructions and performing well-defined tasks, but they struggle with ambiguity and require clear, detailed guidance. Unlike employees who can learn from experience and adjust to new situations, AI agents need explicit training for each task or scenario going forward. It takes us 18 years to build an adult, we cant expect Agents to born and raise in 6 months of prompting. Find a task that creates a lot of pain, and tackle it.
For that task, you need well-documented SOP processes. The foundation of successful AI agent development lies in well-documented processes, often referred to as Standard Operating Procedures (SOPs). SOPs provide a clear roadmap for the agent, outlining the steps involved in a specific task. Starting with SOPs simplifies agent training, as the necessary information is readily available. There are multiple SAP SOP processes, so focus on the platform.
That platform matters a lot. When embarking on your AI agent journey, from an SAP-centric perspective, platform selection is important. The right platform streamlines development, integration, and ultimately, the value your agents deliver. SAP offers AI Core and SAP AI Launchpad which is key if agents must access and interact with SAP data and business processes, but more generalistic frameworks like CrewAI offer flexibility and a growing community, tho they require significant custom integration with SAP systems, data silos, secure connections, and managing the complexities of interacting with SAP’s APIs. No stable platform, no value for the Business
Business will never build their own agents. Bad news for the business but if you work in IT, there is still some hope. The rise of no-code platforms promised to empower everyone to build software, but that makes someone a developer without coding and business doesnt want to become a developer. AI agent platforms and frameworks is hard and complex and requires specialized AI engineers, these are hard to find, systems are complex and unstable like plutonium these days. Building effective AI agents requires a deep understanding of AI principles, prompt engineering, and integration techniques. Platforms may simplify the process, businesses will likely continue to rely on IT to build and maintain their AI agents.
Business also don’t know which agents they need. Business owners may have ideas for AI agent applications, but as it happened with the cloud journey, a preparation is needed, consult is needed, we don’t know what we don’t know. Often, the most valuable automation opportunities lie in areas that may not be immediately apparent. By mapping out the journey and analyzing existing processes, you can identify pain points and areas where AI agents can deliver the most significant impact which might not be the initial plan but the most effective one.
Data-driven decisions with actions/rules deliver results. AI agents are most effective when they can both analyze data and take action based on its analysis. They require domain expertise, and combining data with relevant actions allows agents to execute tasks, make informed decisions, and provide valuable insights. Not only that, the key of Agents will be delivered by a combination of 3 topics, the knowledge they possess, the rules we give them, and the memory they retain. I will go through this in the next points
Tools are the most important component. I just said Knowledge, Rules and Memory, and its true in the brain, but in the tools/function calling, AI agents generate value through the actions they perform, and tools are what enable those actions. Building and structuring tools effectively is crucial for developing AI agents that can solve real-world problems. This goes again the the domain of the platform, this is where SAP is delivering its value against using other platforms, where all tools must be developed from scratch.
Model costs don’t matter, what matters is ROI. Corporations don’t care DeepSeek costs less than OpenAI if the output of the model is irrelevant. That is a problem and a virtue. Prioritize the return on investment (ROI) that an AI agent can deliver. The time and cost savings achieved through automation far outweigh the model costs. As long as the agent delivers on its promises and adheres to data privacy policies, the underlying model is less important.
Enough of use cases, Outcome Based Value. A formula to calculate the OBV could be factors such as employee hourly rates, time spent on tasks, operational costs, or development costs we save by using agents on Joule, this is what industry calls Outcomes based value. Now, how we tie the investment spent on AI agents with the outcomes they produce?
Separate between AI Agents and Agentic AI. Agentic AI refers to autonomous entities that can analyze information and make decisions based on their environment. Workflows follow a predefined sequence of steps, although those steps can be agentic in nature. The key distinction lies in scope and autonomy: an AI Agent is a single, specialized software component designed for specific tasks with limited, pre-defined autonomy, like a chatbot, more related to BTP Generative AI Hub. In contrast, Agentic AI refers to a system or framework where multiple AI agents (and potentially other AI components) interact, coordinate, and potentially even compete to achieve broader, higher-level goals. Agentic AI systems possess a greater degree of system-level autonomy and often exhibit emergent behavior, meaning the system’s overall intelligence and capabilities surpass the sum of its individual parts. It’s the difference between a single worker (AI agent) and a collaborative team with a shared objective (agentic AI). Agentic AI systems such SAP Build Process Automation, focus on orchestrating a workflow, rather than executing isolated actions.
Deploying agents is harder than building them. While building an AI agent is a significant challenge, deploying it into a production environment is way more complex. Again, choosing a platform that offers flexibility, scalability, and ease of integration with existing systems with factors such as security, monitoring, and maintenance is crucial, think the long game.
Include a human in the loop for SAP mission critical agents. Most of the demo AI Agents tasks allow some level of failure, like a travel assistant, SAP tasks come with a 0 margin for error or significant consequences happens because this is ERP, so it’s crucial to include a human in the loop. Joule will be necessary here. You need Joule.
Vertical AI agents are likely to succeed faster than horizontal agents. Vertical AI agents, which specialize in specific use cases or industries, are gaining traction. Just as vertical SaaS solutions have proven successful, vertical AI agents offer the advantage of deep expertise and targeted functionality. Vertical AI agents may succeed more readily than horizontal AI. Horizontal AI agents attempt to be general-purpose assistants, tackling a wide range of tasks across various domains, making it difficult to achieve mastery in any single area. Vertical AI agents concentrate on specific industries or business functions (e.g., SAP-specific Maintenance Notification agent) specialization allows for deeper domain expertise, leading to more accurate models, tailored workflows, and solutions that directly address industry-specific pain points. The data used to train vertical agents is more relevant, the problems they solve are more clearly defined, and the return on investment is easier to quantify, faster adoption and demonstrable success within their niche.
Agents don’t replace people; they help businesses scale. Closing with the opening statement, AI agents are not intended to replace human employees but rather to augment their capabilities and enable businesses to scale. They automate repetitive tasks and processes, free up employees to focus on higher-level activities and leave the humans for human interaction.
I don’t want to interrupt, butafter some time working on AI agents, some factual facts I must share todayAI agents are not your employees and they are not to substitute people, but tasks. AI agents lack the adaptability of humans. They excel at following specific instructions and performing well-defined tasks, but they struggle with ambiguity and require clear, detailed guidance. Unlike employees who can learn from experience and adjust to new situations, AI agents need explicit training for each task or scenario going forward. It takes us 18 years to build an adult, we cant expect Agents to born and raise in 6 months of prompting. Find a task that creates a lot of pain, and tackle it.For that task, you need well-documented SOP processes. The foundation of successful AI agent development lies in well-documented processes, often referred to as Standard Operating Procedures (SOPs). SOPs provide a clear roadmap for the agent, outlining the steps involved in a specific task. Starting with SOPs simplifies agent training, as the necessary information is readily available. There are multiple SAP SOP processes, so focus on the platform.That platform matters a lot. When embarking on your AI agent journey, from an SAP-centric perspective, platform selection is important. The right platform streamlines development, integration, and ultimately, the value your agents deliver. SAP offers AI Core and SAP AI Launchpad which is key if agents must access and interact with SAP data and business processes, but more generalistic frameworks like CrewAI offer flexibility and a growing community, tho they require significant custom integration with SAP systems, data silos, secure connections, and managing the complexities of interacting with SAP’s APIs. No stable platform, no value for the BusinessBusiness will never build their own agents. Bad news for the business but if you work in IT, there is still some hope. The rise of no-code platforms promised to empower everyone to build software, but that makes someone a developer without coding and business doesnt want to become a developer. AI agent platforms and frameworks is hard and complex and requires specialized AI engineers, these are hard to find, systems are complex and unstable like plutonium these days. Building effective AI agents requires a deep understanding of AI principles, prompt engineering, and integration techniques. Platforms may simplify the process, businesses will likely continue to rely on IT to build and maintain their AI agents.Business also don’t know which agents they need. Business owners may have ideas for AI agent applications, but as it happened with the cloud journey, a preparation is needed, consult is needed, we don’t know what we don’t know. Often, the most valuable automation opportunities lie in areas that may not be immediately apparent. By mapping out the journey and analyzing existing processes, you can identify pain points and areas where AI agents can deliver the most significant impact which might not be the initial plan but the most effective one.Data-driven decisions with actions/rules deliver results. AI agents are most effective when they can both analyze data and take action based on its analysis. They require domain expertise, and combining data with relevant actions allows agents to execute tasks, make informed decisions, and provide valuable insights. Not only that, the key of Agents will be delivered by a combination of 3 topics, the knowledge they possess, the rules we give them, and the memory they retain. I will go through this in the next pointsTools are the most important component. I just said Knowledge, Rules and Memory, and its true in the brain, but in the tools/function calling, AI agents generate value through the actions they perform, and tools are what enable those actions. Building and structuring tools effectively is crucial for developing AI agents that can solve real-world problems. This goes again the the domain of the platform, this is where SAP is delivering its value against using other platforms, where all tools must be developed from scratch.Model costs don’t matter, what matters is ROI. Corporations don’t care DeepSeek costs less than OpenAI if the output of the model is irrelevant. That is a problem and a virtue. Prioritize the return on investment (ROI) that an AI agent can deliver. The time and cost savings achieved through automation far outweigh the model costs. As long as the agent delivers on its promises and adheres to data privacy policies, the underlying model is less important.Enough of use cases, Outcome Based Value. A formula to calculate the OBV could be factors such as employee hourly rates, time spent on tasks, operational costs, or development costs we save by using agents on Joule, this is what industry calls Outcomes based value. Now, how we tie the investment spent on AI agents with the outcomes they produce?Separate between AI Agents and Agentic AI. Agentic AI refers to autonomous entities that can analyze information and make decisions based on their environment. Workflows follow a predefined sequence of steps, although those steps can be agentic in nature. The key distinction lies in scope and autonomy: an AI Agent is a single, specialized software component designed for specific tasks with limited, pre-defined autonomy, like a chatbot, more related to BTP Generative AI Hub. In contrast, Agentic AI refers to a system or framework where multiple AI agents (and potentially other AI components) interact, coordinate, and potentially even compete to achieve broader, higher-level goals. Agentic AI systems possess a greater degree of system-level autonomy and often exhibit emergent behavior, meaning the system’s overall intelligence and capabilities surpass the sum of its individual parts. It’s the difference between a single worker (AI agent) and a collaborative team with a shared objective (agentic AI). Agentic AI systems such SAP Build Process Automation, focus on orchestrating a workflow, rather than executing isolated actions.Deploying agents is harder than building them. While building an AI agent is a significant challenge, deploying it into a production environment is way more complex. Again, choosing a platform that offers flexibility, scalability, and ease of integration with existing systems with factors such as security, monitoring, and maintenance is crucial, think the long game.Include a human in the loop for SAP mission critical agents. Most of the demo AI Agents tasks allow some level of failure, like a travel assistant, SAP tasks come with a 0 margin for error or significant consequences happens because this is ERP, so it’s crucial to include a human in the loop. Joule will be necessary here. You need Joule.Vertical AI agents are likely to succeed faster than horizontal agents. Vertical AI agents, which specialize in specific use cases or industries, are gaining traction. Just as vertical SaaS solutions have proven successful, vertical AI agents offer the advantage of deep expertise and targeted functionality. Vertical AI agents may succeed more readily than horizontal AI. Horizontal AI agents attempt to be general-purpose assistants, tackling a wide range of tasks across various domains, making it difficult to achieve mastery in any single area. Vertical AI agents concentrate on specific industries or business functions (e.g., SAP-specific Maintenance Notification agent) specialization allows for deeper domain expertise, leading to more accurate models, tailored workflows, and solutions that directly address industry-specific pain points. The data used to train vertical agents is more relevant, the problems they solve are more clearly defined, and the return on investment is easier to quantify, faster adoption and demonstrable success within their niche. Agents don’t replace people; they help businesses scale. Closing with the opening statement, AI agents are not intended to replace human employees but rather to augment their capabilities and enable businesses to scale. They automate repetitive tasks and processes, free up employees to focus on higher-level activities and leave the humans for human interaction. Read More Technology Blogs by Members articles
#SAP
#SAPTechnologyblog
