AI Agents: Disrupting automation and reimagining productivity

AI Agents are revolutionizing automation, fundamentally altering productivity for businesses and knowledge workers alike. Experts discuss the transformative impact of AI on automating complex workflows, enhancing human-in-the-loop systems, and creating new efficiencies across industries.

Key speakers

Moderator: Praveen Akkiraju, Insight Partners Managing Director
Speaker: Ruchir Puri, IBM Research Chief Scientist
Speaker: Daniel Vassilev, Relevance AI Cofounder and Co-CEO
Speaker: Artem Harutyunyan, Bardeen AI Cofounder and CTO

These insights came from our ScaleUp:AI event in November 2024, an industry-leading global conference that features topics across technologies and industries. Watch the full session below:

Key takeaways

• AI Agents are evolving beyond copilots to autonomous systems capable of executing complex tasks with minimal human input.
• The shift from “feed-forward” AI systems to feedback-driven Agents enables greater reasoning, planning, and adaptability.
• Specialization and domain-specific expertise are critical for Agents to function effectively in real-world AI applications.
• AI-driven automation is transforming key business functions, particularly in software development, sales, procurement, and HR.
• Human-in-the-loop remains essential for trust and oversight, but as AI Agents prove consistent and reliable, dependence on human supervision will decrease.
• The economic model for AI agents is shifting from subscription-based pricing to outcome-based pricing, aligning costs with tangible business impact.

Defining AI Agents: From copilots to full autonomy

The term “AI Agent” is widely used but lacks a universally agreed-upon definition. According to Harutyunyan, an AI Agent should function as a system where users can explain their goals as they would to a human, and the Agent autonomously determines how to accomplish them.

Vassilev framed AI Agents through the lens of automation, distinguishing between copilots and autopilots. “A lot of people are familiar with copilot-based assistance, like ChatGPT,” he explained. “But the thing that I that I try to communicate when we talk about Agents, and what I think is really the value that they kind of tap into, is the autopilot piece. It’s all about doing work and delegating work to an Agent. Can it do it on autopilot, autonomously?”

“It’s all about doing work and delegating work to an Agent. Can it do it on autopilot, autonomously?”

Puri took a technical approach, emphasizing the shift from traditional AI systems to feedback-driven Agents. “AI Agents bring AI systems from a primitive era to an era where each of us humans are. We are all feedback systems,” he said. Unlike static feed-forward models, Agents operate in feedback loops, continuously improving outcomes. “We can label it anything, copilot, or whatever we want, but the technology underneath it speaks for itself. Once you get to these feedback systems, they are amazingly powerful.”

The core technologies enabling AI Agents

The discussion then turned to the key technologies that enable AI Agents to function effectively. Puri highlighted four critical components: domain specialization, tool usage, execution environments, and verification. “Grounding and specialization are key to success,” he said. “The real question is, do [Agents] reason in a certain domain?” Puri explained the four steps to building successful Agents:

• Grounding and specialization: Rather than making AI Agents overly generic, they should be designed to plan and reason effectively within specific domains. The real question isn’t whether models can reason but whether they can reason within a given domain.

• Tool utilization: AI models must learn to call the right tools at the right time. For example, ChatGPT struggles with simple arithmetic because it’s not a prediction task. Instead, it should call a calculator when needed, demonstrating the importance of function calling.

• Execution environment: AI Agents need an environment where they can execute tasks properly, ensuring they follow through on their planned actions.

• Verification and reflection: The most critical factor for AI Agents is their ability to verify and check their own consistency. Effective reflection allows them to refine their processes, making enterprises that prioritize consistency checks more successful.

Real-world applications of AI Agents

AI Agents are already being deployed across various industries. Vassilev highlighted the importance of designing AI workflows based on domain expertise. “Our goal has always been to help companies deploy an AI workforce, and for that to do its job, it has to be robust, reliable, and accurate,” he said. “Step one in that process is not actually thinking about the technology…it’s the process. What is the unique workflow that your business does for that specific process that makes it successful?”

Puri outlined three key areas where AI Agents are gaining traction: software development, sales and procurement, and HR. “HR domains, because there are workflows, processes, tools. And if you get it wrong, there are implications to it, right?”

“When will you not have a human in the loop?”

Despite advancements in AI autonomy, human oversight remains a critical factor. “Automation is all about trust,” Puri stated. “When will you not have a human in the loop?”

Harutyunyan provided an example from recruiting. While recruiters may believe their work isn’t repetitive due to the uniqueness of candidates, hiring managers, and roles, much of their time is spent on administrative tasks like data entry, emails, and reminders. The core intelligence required for quick candidate assessments is difficult to replace without advanced AI, but automating the mundane workflow — such as transferring data between platforms — can dramatically increase efficiency. The goal isn’t to replace humans but to enable them to accomplish significantly more while focusing on meaningful, high-value work.

Vassilev expanded on this by comparing AI Agents to junior employees. “We think it’s very similar for AI Agents. These are junior employees that are joining your organization today, and we so we build in escalation pathways so that when the Agent doesn’t know how to do something, it can request a manager to help it or human to help it.”

The changing economics of AI Agents

The panel also explored the shifting economic models around AI adoption. Historically, software has been priced through licensing or subscriptions. However, AI agents enable a move toward outcome-based pricing.

“Agents bring us closer to the outcomes.”

Puri described this shift as a “tug-of-war” between old and new business models. “Agents bring us closer to the outcomes. And it will disrupt some business models,” he explained. “I definitely believe we as consumers will start to [look to] more of outcome-based pricing as we continue…because software-driven Agents bring us closer to that outcome than ever before.”

Harutyunyan predicted that as AI technology advances, the cost of intelligence will approach zero. “In the future, the marginal cost of intelligence will follow a similar trajectory,” he said. He compared the decreasing cost of AI intelligence to how data storage costs have approached zero over time.

While AI processing today can still be expensive, he anticipates that within a few years, costs will drop, making AI essentially free at the margin. On the supply side, this means AI will become far cheaper to produce and deploy. On the demand side, however, AI products are currently priced based on their outcomes rather than production costs, leading to high prices.

Watch more sessions from ScaleUp:AI, and see scaleup.events for updates on ScaleUp:AI 2025.