Operationalising AI Agents - with Dr Sokratis Kartakis

In this episode of Humbot, we dive deep into what it really takes for organizations to move from experimental AI pilots to full-scale, production-ready agentic systems. Our guest is Sokratis, GenAI Black Belt at Google, who has helped some of the largest enterprises in EMEA and APAC operationalize GenAI and AI agents.

Key Takeaways

Start with business value – Education comes first

Define ROI and identify real process bottlenecks before building tech. Understanding the business context and value proposition is crucial for successful AI agent implementation.

Foundations matter

Set up a secure landing zone, establish a governed data lake, and ensure access management before scaling agents. The infrastructure foundation determines the success of your AI initiatives.

Memory & RAG

Agents need short-term and long-term memory, plus centralized retrieval-augmented generation for multi-use cases. This enables agents to maintain context and provide relevant, accurate responses.

Actions done right

Use MCP to standardize tool calls, and A2A to let agents from different frameworks collaborate seamlessly. Proper action orchestration is key to agent effectiveness.

Enterprise security

MCP is evolving, but expect managed services from hyperscalers soon. For now, layer in strong governance and access control to ensure secure deployments.

Evaluation is ongoing

Combine offline evaluation, guardrails, and real-time monitoring. Metrics should cover infrastructure, outputs, tool calls, and memory relevance for comprehensive assessment.

People first

AI agents don't eliminate jobs—they transform them. Upskill your teams in prompting, governance, and evaluation to maximize the benefits of AI integration.

Startups: specialize

Success comes from solving vertical problems with GenAI, not from reinventing infrastructure. Focus on domain-specific solutions rather than building generic platforms.

Why It Matters

AI agents are no longer a "lab experiment." With the right foundations—infrastructure, data, governance, and security—businesses can unlock real ROI, scale across use cases, and ensure safe, reliable deployments. As Sokratis notes, the future is not just about chatbots but about ubiquitous, multimodal agents woven into customer support, internal workflows, and even industry-specific verticals.

Expert Insights

"Start with business value—education first, then technology." – Sokratis

This fundamental principle guides successful AI agent implementations. Organizations that prioritize understanding their business needs and educating their teams before diving into technical solutions are more likely to achieve sustainable success.

The Path Forward

The conversation reveals that enterprises are already using MCP and A2A in production, demonstrating that the technology is mature enough for real-world applications. However, success requires:

• Strategic Planning: Understanding where AI agents can provide the most value
• Technical Foundation: Building robust, secure infrastructure
• Team Preparation: Training teams for the future of AI-human collaboration
• Continuous Improvement: Ongoing evaluation and optimization

Video Highlights

Here are key moments from our conversation with Dr. Sokratis Kartakis:

Clip 1: Securing AI Agents at Scale

Clip 2: Unifying AI Agents: The Role of MCP and A2A in Driving Business Results

Clip 3: New Agentic AI Job Categories: Shifting The Conversation from Threat to New Opportunities

Clip 4: Scaling AI Agents: Mastering Evaluation for Deployment

Conclusion

🎧 Listen to the full episode to hear how enterprises are already using MCP and A2A in production, why memory and RAG matter more than ever, and how to prepare your teams for the future of AI agents.

The insights from Dr. Sokratis Kartakis provide a roadmap for organizations looking to move beyond AI experimentation to production-ready, scalable agentic systems that deliver real business value.