How Do You Make AI Agents Reliable at Scale?

At Big Data LDN (BDL) 2025, Keboola CEO Pavel Dolezal presented a new data agent designed for all business users, not just engineers. With a mission to make AI, automation, and data easy to access, relevant, and useful across the organisation, Dolezal revealed that the data agent has been embedded with contextual intelligence and generative AI.

“While we typically assist data engineers with building the pipeline, we took the same data agent and built a different environment for it — a chat-like environment. By default, the chat has context, knows what to do, knows where not to go,” the Keboola CEO unveiled on the Don’t Panic It’s Just Data podcast.

In the EM360Tech podcast recorded live at BDL, Dolezal spoke to Christina Stathopoulos, the Founder of Dare to Data, in the recent episode of the Don’t Panic It’s Just Data podcast. They talked about the new Keboola Data Agent and it plays a key role in AI-backed change and the growth of large language models (LLMs) in business.

Context for AI-Backed Data Strategy Matters More

“Anyone can be technical now. It’s context that matters,” stated Dolezal, also the co-founder of Keboola. He presented a strong argument for why enterprise data strategies are falling short and how a new wave of smart tools will change that.

“The pipeline of what you can do is limitless if you build it for people in business,” he added. “You can't keep data and AI just in the hands of engineers anymore. That model doesn’t scale.”

As businesses face a growing number of data sources — sometimes over 300 SaaS platforms and more than 80 departments — managing, governing, and activating that data has become a challenge. The appealing promise of AI often adds another layer of complexity.

When AI Adds More Complexity

Enterprise leaders were told that AI would simplify data workflows. Instead, many found themselves managing disconnected tools and failed pilot projects.

“We all read the MIT study. 95 per cent of AI proofs of concept don’t make it to production,” the CEO of Keboola highlights. “Why? Because large language models (LLMs) need context. And enterprise data environments are anything but simple.”

At Keboola, context is crucial, he emphasised. It includes not just metadata but also event logs, debug trails, and orchestration details, including the complete story behind every data product.

“LLMs thrive on context. The more relevant context you provide, the better the outcome. But in today’s data stack, where your context is spread across 15 tools, that's nearly impossible.”

This is where Keboola’s new Data Agent comes in. It is a generative AI interface built directly into data workflows, capable of understanding and acting on both the structure and state of a company’s data.

Watch the podcast for further insights on EM360Tech.

Key Takeaways

Focus on context and domain knowledge rather than technical skills.
Ease of acquiring technical skills compared to the importance of understanding business processes.
Agents will run business processes both internally and externally.
Need for infrastructure to support agents
Challenges of provisioning ad hoc environments
Use agents to automate business processes.
Importance of data governance.