Until the data is ready, the agents won't be either

CEOs and the board want agentic AI yesterday. But as Informatica's Steven Seah puts it, agents are only as good as the data they are given.

Unfortunately, that data is messy right now. Here's also what I am hearing from actual AI and chief data officers in the room.

AI is a sequential journey

You can't just skip straight to AI, observed Steven at the 9th CDDO Summit today. In fact, many customers try to make the leap to AI without first dealing with their data, and end up stuck.

In his presentation, he shared a three-stage journey: unlocking data across silos, building trust through governance and validation, and activating it to power agents, workflows, and analytics.

His message is blunt. Until organisations treat context as a first-class system rather than an afterthought, most AI pilots will struggle to deliver real outcomes.

Easier said than done

Deploying AI in the enterprise takes more than subscribing to Copilot or another AI chatbot. I had the opportunity to hear from several AI and data practitioners on this. Here are my personal takeaways.

• AI isn't magic. The challenge today isn't so much that company executives don't believe in AI. It's when some think of it as "magic" that can be rolled out overnight, or next month. Another issue is when senior leadership hasn't clearly thought through the work, or cost, of rolling out AI fully. This can result in a "cha-cha" dance as stakeholders waver.
• AI accelerates the problems. Think you have data problems such as silos or data of questionable quality? AI deployments will simply expose them faster. And if your data permissions are fragmented, be prepared for agents to unwittingly pick them up at AI speed, with potentially disastrous consequences.
• Mandate not optional. A mandate right from the top can make all the difference to the successful implementation of AI. This means the C-suite must want it, and preferably at the board level too.

Conclusion

It was fascinating to hear from AI and data professionals today. From finding business value to rushing out new AI governance standards, everyone is wrestling with a very different set of challenges.

But one thing is clear. Until the data is ready, the agents won't be either.