From: aidotengineer
Hubert Mishtella, who leads AI researchers in drug design at Novardis, presented on “Agentic Enterprise: What Your CEO Must Know About AI” at the AI Engineer World’s Fair 2025 [00:00:00]. The presentation covered various aspects of AI including workflows, networks, challenges, and the individual employee’s perspective on how AI agents might be used and their organizational impact [00:00:21].
The Rise of AI Agents in Enterprise
It is suggested that within three years, organizations might be run by AI agents [00:00:50]. An AI agent is a large language model (LLM)-based application that exhibits autonomy in its behavior [00:01:08]. They are composed of a language model that connects various tools, executing tasks through step-by-step planning [00:01:22]. The use of these tools can be dynamically adjusted based on the outcome of previous steps, known as dynamic action [00:01:37].
Key Capabilities of AI Agents
AI agents introduce capabilities previously almost unique to humans [00:01:47]:
- Reasoning for multi-step planning [00:02:26]
- Adaptability for dynamic work control [00:02:28]
- Persistent memory [00:02:32]
- Contextual knowledge (Retrieval Augmented Generation - RAG): Utilizing specific knowledge not available in the language model [00:02:35].
- Interactive workspaces: Collaborating with an AI agent in sandboxes or canvases to refine tasks like code or art [00:02:49].
- Computer-using agents: Navigating graphical user interfaces (GUIs) to automate steps previously done by humans or robotic process automation [00:03:02].
Many companies are already working on evolving their digital assets for AI agent usage, aiming for every digital asset to become a tool for an AI agent [00:01:55].
Workflow Perspective: Automation and Augmentation
A workflow is the execution of a process, often involving sequential steps [00:03:32]. Technology can be used in two primary ways:
- Automation: Substituting a specific step with technology [00:04:15].
- Augmentation: Performing a step quicker or better [00:04:26].
While machine learning (ML) could previously automate or augment some tasks [00:04:31], AI agents offer a significant difference by tackling a broader set of tasks that operate on cognitive steps, often represented in natural language [00:05:07]. They also act as a “glue” between tasks, allowing for their composition without human intervention, except for feedback when needed [00:05:20].
Agentic Workflow
The concept of an “agentic workflow” describes a situation where AI agents automate and delegate not just one, but several steps in a process [00:05:39].
To begin building agents, organizations should first identify their workflows and understand the exact context surrounding them [00:05:55]. This context—including data, systems, transformations, and steps—is often only in an employee’s head and not documented [00:06:11].
Organizational Impact: Network of Personas
Relying solely on traditional roles and job titles is insufficient for applying AI agents effectively [00:07:50]. Understanding the exact tools to use, planning, and conditions for behavior change requires deeper insight than job descriptions provide [00:08:00].
A “network perspective” suggests thinking about each contributor to a workflow as a distinct “persona” [00:08:28].
- Silent Achiever: Lower communication, high performance [00:08:57].
- Individual Contributor: Both communication and performance [00:09:04].
- Connector: Links non-adjacent teams [00:09:09].
- Multiplier: Enhances the work of other team members [00:09:16].
- Knowledge Hub: Provides domain expertise [00:09:21].
Defining these personas helps project how AI agents might impact workflows and specific employees [00:09:31]. For example:
- A “silent achiever” with a coding assistant might become so productive they can handle work previously done by multiple people or steps [00:11:18].
- An agent can magnify a “multiplier” persona’s impact across all team members, not just adjacent ones [00:11:56].
- “Knowledge hub” personas might be most prone to substitution by an AI agent that dynamically supports the entire team [00:13:03].
This approach allows for identifying repeating patterns across workflows, leading to optimized agent development [00:13:55].
Emerging Patterns and New Roles
AI will lead to shifts in task execution [00:14:20]:
- Human Tasks: Ambiguous workflows where outcomes depend heavily on the situation, or where subjectivity is desired, are better executed by humans [00:14:42].
- AI Agent Tasks: Repetitive, tedious tasks that require perfect, consistent performance are ideal for AI agents [00:15:12].
Emerging patterns include:
- Commoditization of Intelligence and Domain Knowledge: These become cheap and easily available, no longer a unique edge for employees or companies [00:15:32].
- Shift to Multi-disciplinary Know-How or Deep Specialization: Employees and companies will need to pivot to gain new advantages [00:16:03].
- AI as Super Connectors: Agents can facilitate communication and impact across teams [00:16:43].
- Increased Value of Context Understanding: This will gain more value than intelligence or domain expertise alone [00:16:48].
- AI Supercharges Doers: Individuals who perform tasks can do even more and quicker, potentially becoming “100 times engineers” [00:17:04].
- Polarization of Positions: Less “middle ground,” with a push towards new specializations [00:17:27].
New roles are expected to appear in AI-mature organizations [00:17:42]:
- Workflow Minor [00:17:48]
- Human-AI Orchestrator [00:17:55]
Democratizing AI Agents
To enable organizational adoption of AI, a step-by-step approach is crucial for democratization [00:18:15]:
- Access to general assistants: Already common in many companies [00:18:30].
- Individual assistants per employee: With memory, adjustments, context understanding, and personal databases [00:18:39].
- Employee-built agents: The key skill will be employees building agents on the spot using coding assistants, no-code, or low-code tools [00:19:10]. This requires cognitive self-awareness to translate mental steps into specific, automatable actions [00:19:45].
- Digital Twins: Agents representing employees, retaining organizational knowledge, allowing “time travel” to reconstruct past expertise [00:20:24].
- Agents for multiple employees: Serving entire teams or workflows [00:21:06].
- Swarm of agents: Already becoming a reality with maturing LLM and AI agent software stacks [00:21:12].
Challenges and Opportunities
Challenges
- Technical: Groundedness, hallucinations, guardrails, security [00:21:42].
- Adoption: The shift from humans driving interaction to AI agents driving interaction and delegating subtasks to humans poses a significant cultural question for organizations [00:22:15].
- Ethics of Autonomy: This raises questions about what values AI agents should store and how to resolve conflicts between human and AI agent actions [00:22:59]. The greater a machine’s freedom, the more it will need moral standards [00:24:12].
Opportunities
AI offers a list of interesting opportunities for executives [00:24:44]:
- Simulating synthetic reality before real-life activities (e.g., marketing, lab work) [00:24:53].
- Proactively understanding regulatory loopholes [00:25:06].
- Finding merger and acquisition (M&A) opportunities [00:25:09].
These developments necessitate more discussions around philosophy, ethics, social aspects, and psychology of AI and humans [00:25:30].
Executive Summary
For CEOs, several points are crucial regarding AI in workflow automation and augmentation [00:26:27]:
- If a company’s strategy still makes sense after swapping “AI agent” with “machine learning” or “data,” it is likely outdated [00:26:29].
- AI agents offer new capabilities that make a significant difference [00:26:56].
- A detailed understanding of workflows is crucial to maximize benefits from agents [00:27:00].
- Democratizing agents is essential to fully accelerate the AI agent revolution [00:27:05].
- Considering the network of human personas is highly important [00:27:11].
- Organizations must be ready for rapid transformation, as entire organizational charts might change drastically [00:27:18].
- A mindset shift is essential, preparing for interactions where technology might even lead [00:27:31].
- Ethics remain a crucial part of this transformation [00:27:47].