Agentic AI: More “Personal Chef” Than “Vending Machine”

In an exclusive interview, Sumeet Agrawal, VP of Product Management at Informatica, sheds light on the transformative potential of Agentic AI. Agrawal defines Agentic AI as a significant leap beyond traditional rule-based systems, likening it to a “personal chef” capable of understanding context and making intelligent decisions, rather than a rigid “vending machine.”

How do you define Agentic AI, and how does it fundamentally differ from traditional chatbots or rule-based automation?
Agentic AI represents a significant leap forward from traditional chatbots or rule-based systems. While the latter are confined to rigid sets of instructions, they falter when faced with ambiguity or unfamiliar scenarios. They’re like vending machines: push the right button, and you’ll get exactly what you asked for, nothing more, nothing less. But try requesting a customised dish or handling a last-minute change? They simply aren’t built for that.

Agentic AI, on the other hand, operates more like a personal chef. Powered by large language models (LLMs), it can understand context, infer intent, and make intelligent decisions to achieve a broader goal. It can navigate complexity, plan its next move, and adapt to the unexpected. That’s what makes it agentic: the ability to act with a degree of agency, rather than simply following instructions.

In essence, the difference lies in flexibility and intent. Rule-based systems are excellent at completing narrow, structured tasks, but Agentic AI is designed to operate in dynamic environments, making it far more capable of solving novel or evolving problems without needing to be micromanaged every step of the way.

Beyond task automation, what complex, multi-step problems are agentic AIs uniquely positioned to solve that current AI models cannot?
What truly sets Agentic AI apart is its ability to operate with a level of autonomy that closely mirrors how humans approach problems. Rather than relying on a fixed set of instructions, it can grasp a broader objective and chart its own path toward achieving it, even when no pre-defined steps exist. It’s capable of building a plan, taking action, and continuously learning from real-time feedback.

Crucially, Agentic AI can assess whether each step is working, adapt if things go off course, and intelligently shift direction when new information comes to light. By moving beyond linear, rules-based task execution, Agentic AI unlocks new possibilities for solving problems that previously required constant oversight. These are real world challenges that typically involve ambiguity, adjusting on the fly, and driving outcomes in a far more human-like way.

How much autonomy should Agentic AI have in decision-making, and where should humans remain in the loop?
It’s best to view autonomy for Agentic AI on a sliding scale, tailored to the specific task at hand. It’s not a matter of either handing over the reins entirely or keeping AI on a tight leash. It’s about calibrating the balance of control and oversight based on the impact and risk involved.

For routine, low-risk tasks such as summarising meeting notes or updating internal dashboards, Agentic AI can act almost independently. In these cases, speed and efficiency matter more than precision or oversight. But as the stakes rise, so too should human involvement. High-impact decisions, especially those involving legal, ethical, or reputational risk, demand more stringent human oversight. The same goes for situations where the AI’s confidence in its own output is low or where compliance mandates come into play.

The goal is a true partnership. Agentic AI can be leveraged to handle the heavy lifting, analysing data, exploring options, even making preliminary decisions, while humans step in to provide judgment, nuance, and accountability when it matters most. This approach ensures we harness the full power of AI without compromising on trust, safety, or strategic direction.

What infrastructure challenges (such as power, cooling, latency) arise when deploying Agentic AI at scale?
Scaling Agentic AI for enterprise use demands massive infrastructure. This includes addressing enormous power consumption and cooling requirements due to intensive processing. Latency is also a concern, as multi-step reasoning and system interactions can slow responses. Furthermore, significant computing power, memory, high-speed networking, and sophisticated orchestration are essential. Hyperscalers and enterprise software vendors, like Informatica, are key in encapsulating & simplifying the deployment of these complex systems, thereby enabling customers to scale seamlessly.

What’s the next frontier for Agentic AI—will we see AI “agents” collaborating like human teams?
Multi-agent AI systems are already here so the next step is to elevate their collaboration to be far more profound, autonomous, and adaptive, in essence, mirroring skilled human teamwork.

Looking to the near horizon, the real magic will happen when AI agents start working together more like skilled human teams. Instead of just exchanging structured data, they’ll be able to build shared understanding, negotiate when needed, adapt on the fly, and even explain their thinking to each other and to us.

We’ll also see these AI teams becoming far more autonomous, learning from every interaction, refining how they work together, and becoming better collaborators over time. You’ll be able to give these teams a high-level goal, and they’ll take it from there – dividing up the work, assigning roles, and integrating everything seamlessly to deliver a well-rounded result.

Essentially, AI agent teams will evolve into dynamic, self-learning expert systems capable of tackling complex problems with progressively less direct human intervention.

Will open-source models keep pace with proprietary Agentic AI systems, or will there be a widening gap?
I don’t believe it has to be one or the other. Instead, the future of Agentic AI will see both open-source and proprietary models playing distinct, vital roles. As has been the case in the software industry, open-source Agentic AI will likely act as a community-driven R&D lab, excelling in rapid innovation, collaborative breakthroughs, accessibility, transparency, and deep customization. It will continue to advance core capabilities and novel architectures.

In parallel , proprietary Agentic AI will functions like a commercial factory, focusing on delivering reliable, scalable, and enterprise-ready solutions. The strengths of these systems lie in robust governance, security, compliance, scalability, ease of use for businesses, and integration.

So, while open-source will drive foundational innovation, proprietary systems will likely lead in polished, supported enterprise packages. This isn’t necessarily a widening gap in core intelligence but a difference in focus: open-source pushes possibilities, while proprietary solutions enable practical, secure business adoption. A symbiotic relationship, with proprietary systems often building on open-source foundations, is probable.