I spent last week in San Francisco at the Dreamforce conference, attending and absorbing sessions on the myriad ways agentic AI is redefining what is possible in the digital world. It was an energizing experience to hear from the experts – but it was also a reality check. As I listened to speaker after speaker, I realized that, for all of AI’s promise, success is not a guarantee.
It is now crystal clear to me that many AI projects, to borrow a quote, are in “AI purgatory.” The power is real. The value is great. But here we are, suck. Endless pilots, endless talk.
The hard truth is this – without data, Agentic AI is nothing. Weak—a parlor trick. With data, agentic is everything. Truly a new era of digital transformation.
I can hear the heads hitting the table now. “Of course it’s about the data… duh.” Let me unpack this; words matter and how we use them matters. So how we get to the kernel of this hard truth requires some precision around the word “data.”
But let’s first start with the four truths of agentic.
The power of agentic AI is directly correlated to the data the agent has available to use. The informational context that the agent can “see” matters. Why? Because….
AI agents are non-deterministic. Their power is derived from their ability to mimic human thinking, not following deterministic “if/then” logic. This ability to change and evolve in a human-like manner is what sets them apart. But for agents to thrive we must understand that…
Not all data is the same. Just like with human knowledge, some information (data) is easier to “see” than other information. This is something we understand naturally as humans—we only have the information we have been exposed to. So, like a person, AI agents need to be exposed to the data they need to “see.” And beyond the “seeing” is also understanding that…
Data context is king. Just having access to “see” data is not enough; the context of the data matters—a lot. Humans intuitively understand data and information in context to such an extent that we view them as inseparable. But they are, in fact, two separate facets of understanding.
Context is King
Let’s play a quick game. Pretend you are messaging someone. They don’t know you. You don’t know them. (And yes, this is a variation of the Turing Test.)
You say: “I have a headache.” This is a piece of data.
Without context, it’s basically meaningless to the other person.
The receiving person would reply something like “???”
Let’s add some context. The message is to a doctor.
The data (“I have a headache”) and the context (“the message is to a doctor”) combine to create understanding. The doctor would infer that you are asking for help, and you would be expecting help.
The reply would something like: “I would recommend taking Advil.”
Let’s add more context. The message is to a doctor you have seen for 20 years.
This combination of data and context is now much more powerful. The doctor now “sees” all your data (medical records, visits, lab tests, etc.).
Now, this same data (“I have a headache”) is now many times more relevant to you because of the rich context. The reply would be something like: “You need you to go the ER now.”
Why? The combination of data and context matters. Your doctor knows you were recently in a car accident. Your doctor knows this is urgent.
The point is clear. Context is king.
And for AI agents, we need to understand that their context comes from data. After all, they are still computers. Intuiting context is natural for humans, not for AI.
(This process is often referred to as personalization — and this is what we mean by personalization: Data and context working together, creating relevance to you.)
Data, Data Everywhere and Not a Drop to Use: What We Mean When We Say “Data”
For AI agents to deliver, we humans must create the data environment (the context) in which they can thrive.
At real world level this means a bit of a shake up to the typical IT thinking.
If you are an IT leader, your new mission is not to create deterministic outcomes — computers are already great at that. Your mission is to create a data environment (the context) in which AI agents can thrive, adapt, and evolve in a non-deterministic way, maximizing agentic AI’s ability to adapt as facts and data change.
What Does Creating a Data Environment for AI Agents Mean for IT Leaders?
Collect all the data. Collect it even if you’re not sure yet exactly how it might be used. Do this from every system. This does not mean replacing any or all systems, but it does mean having a data strategy that ingests data from every existing system and makes it available to AI agents—and humans.
Harmonize the data. Different systems identify people, events, etc. in different ways. You need to develop unified profiles to overcome these differences. For example, if system A Identifies a person via email address and system B uses a social security number, your internal data collection needs to be able to understand that these interactions are the same person. As with above, this is not a full-on system change – it's developing and implementing data strategy.
Create a semantic layer. Like us humans, AI agents thrive when the meaning of the data is clear. While we do this naturally, we need to give our AI agents a bit of a helping hand. A semantic layer helps our AI agents to understand the deeper meaning of the data, not just the bits and bytes.
For example, the semantic layer helps AI agents understand that when you say, “how many issues were resolved last week” you mean from any system, look for words like “help, trouble, case, problem, issue.” For humans, the concept of issues would naturally include those related terms – you would not even need to explain to another human that these concepts are related. But because AI agents don’t have that intuitive understanding, a semantic layer clarifies the relations between concepts. This is especially true in industries where there is a lot of specialized language/jargon.
Develop a governance ontology. Data needs to be tagged and, in some cases, masked from AI agents, to make clear what is off limits or can only be used in specific contexts. Again, this is natural for humans, but AI agents need a bit of help understanding the context around what can be used—and when, how, and where.
Conclusion
So, when I say AI agents need “data” all of this is what I mean. It's not the head-to-table obvious statement it may have sounded like at first. It’s a transformation of the data strategies that have ruled for the last 25+ years.
At AgencyQ, we have a team of data experts ready to help you make this transition. We’re ready to help you create a data environment where your AI agents will grow and prosper—and so will your business! Contact us today to get started!