Panel - AI Agents & Intelligent Automation: The Next Evolution of Workflows
Bartosz Schneider, Shanshan Hou, Paul Chang
AI agents, unlike traditional AI tools, operate autonomously across multiple systems, enhancing productivity and collaboration. They execute workflows, adapt in real-time, and learn independently. Despite their potential, challenges like data silos and integration bottlenecks hinder deployment. Trust and transparency remain critical, especially in sectors like healthcare. Organizations should approach AI integration cautiously, focusing on manageable tasks, ensuring data accessibility, and fostering user adoption while balancing autonomy with explainability. Effective implementation requires strategic planning and understanding user needs.
"Make a plan for how to manage adoption of the automation you want to deploy. One thing that I've also seen is that some people are just not as excited as I am about the things I'm building. And then you get users where you say, please try this new thing I've built, a thing that will make your life better."
Summary
- AI agents can act autonomously, adapt, learn, and make decisions, reshaping productivity and collaboration. - Panelists discussed moving AI agents from concept to deployment and the challenges involved. - Coding and automating information processing are key areas where AI agents excel. - Trust and transparency are crucial when designing AI agents, especially in sensitive sectors like healthcare. - Organizations should not be overly ambitious and should plan for gradual integration of AI agents.
Article
AI agents: the quiet revolution reshaping how we work
Panel of experts debates the future of autonomous AI systems at Helsinki tech summit
In a world increasingly fascinated by artificial intelligence, a more subtle revolution is taking place beyond the headlines about chatbots and image generators. AI agents – autonomous systems capable of executing complex workflows without constant human supervision – are poised to transform how organisations operate, experts revealed at a recent technology conference in Helsinki.
At the VERGE AI Frontier summit held in Helsinki on April 14, 2025, three industry specialists explored the evolution of these systems during a panel titled "AI Agents & Intelligent Automation: The Next Evolution of Workflows." Unlike traditional AI tools that require continuous direction, these agents represent a fundamental shift toward systems that operate across multiple tools, adapt in real time, and learn autonomously.
Beyond reactive systems
"AI agents represent the next generation systems that can take action on our behalf and operate across multiple tools, adapt in real time, and even learn autonomously," explained Bartosz Schneider, principal data strategist at Supermetric, who manages the company's internal automation strategy.
The panel, which also featured Shanshan Hou from Thermo Fisher Scientific and Paul Chang, a machine learning engineer at Datacrunch, grappled with a central question facing the industry: "How do we design and scale systems that are no longer reactive, but increasingly proactive?"
Chang, who holds a PhD in machine learning from Aalto University, traced the technological trajectory from basic chatbots to today's more sophisticated systems. "Everyone's used ChatGPT; initially, it was like a better version of Google. But then we got into a loop of copying and pasting, which was laborious. Moving towards automated workflows was the first step," he said.
Infrastructure challenges remain significant
Despite their potential, the panellists acknowledged that significant obstacles remain before AI agents become mainstream. According to research cited during the discussion, most failed AI adoptions break down not at the model level but in the underlying infrastructure.
Schneider highlighted how traditional data warehousing approaches are often inadequate for supporting agent-based systems. "Traditionally, data warehouses centralised data for analysis. Now, ecosystems are changing," he said, describing how his team had pivoted from centralising data to accessing it live for applications, such as extracting insights from call transcripts and customer relationship management systems.
In healthcare and other sensitive industries, the challenges are even more pronounced. "In healthcare, data privacy is a bottleneck," Hou noted. "We need safer ways to use private data."
Trust and transparency: the non-negotiables
As a user experience designer with a background in neuroscience, Hou emphasised that trust remains foundational to successful AI agent implementation. "Performance is foundational for trust. We must clarify agents' limitations. In healthcare, AI suggests but doesn't diagnose," she explained.
The panel discussed practical approaches to building trust, including providing traceability and transparency with progress indicators, reasoning models, and ensuring users maintain ultimate control. The consensus was that organisations must carefully balance automation capabilities with explainability – particularly in high-stakes environments.
Practical advice for organisations
The experts offered pragmatic guidance for businesses looking to implement AI agents. "Try not to be too ambitious because the general agents or AGI, they are not coming yet. Maybe treat this agent as like a new role to help the existing roles to think smarter or make better decisions," Hou advised.
Chang suggested breaking tasks into simple graphs and using a tiered approach to model deployment: "Use cheaper models for simpler tasks and expensive ones for reasoning."
Meanwhile, Schneider emphasised the human element in successful deployments: "Manage adoption. Not everyone is excited about new technology. Consider customer adoption and how to integrate AI efficiently."
Areas of immediate impact
The discussion highlighted specific domains where AI agents are already delivering significant value. Coding emerged as a prime example, with Schneider sharing his own experience: "I'm a data analyst, not a coder, yet with LLMs I transitioned to coding. These tools transform non-coders into proficient ones."
Automating information processing in language-encoded data was identified as another area showing immense potential, though the panellists agreed that tasks affecting human safety should remain primarily under human control.
As businesses and organisations look toward a future increasingly shaped by autonomous systems, the Helsinki panel provided a grounded perspective on both the transformative potential of AI agents and the careful, measured approach required to deploy them successfully. The revolution may be quiet, but its impact on how we work appears increasingly profound.
Part of VERGE | The AI Frontier: Creativity, Security & Collaboration