Olly Warren - Technical Consultant

You're gonna have to be more specific

Why AI needs context before it can become useful enterprise intelligence

I want to talk about context. Not context as a technical concept. Not just prompt engineering. Not just better instructions.

I want to talk about context as the thing that makes communication, decisions and critically, AI agents actually work.

I think a lot of people still interact with AI as if it can read their mind, and, as I'll come onto, poor responses are mildly annoying when AI is just answering questions, but it becomes much more serious when AI has tools, access and permission to act.

An average evening in my house

I want to start by introducing my Wife - She's not here - which is probably good as I'm about to use her as an example.

Quite often, on an evening, I'll get something like this from her:

Who was that guy?

That's it. No setup. No context. Just: "Who was that guy?"

And my response is usually:

You're gonna need to be more specific

Then starts the interrogation:

And we've arrived at the answer. But, only because we slowly rebuilt the missing context.

The Problem

The important thing here is that she wasn't necessarily asking a bad question. In her head, the context was obvious. She could see the film, she knew the scene, she knew what actor she meant. She had the whole chain of meaning in her head, including what prompted the thought in the first place.

But I didn't have any of that. I just got "Who was that guy?".

Without the context, how was I supposed to know what she was talking about?

That's the point I want to start with: a lot of communication only works because of shared context. When context is missing the other party has to guess, clarify or infer.

AI gets the same treatment

This is exactly how a lot of people interact with AI. They ask something vague, with a fully formed expectation in their own head and then get frustrated when the AI doesn't produce the thing they imagined.

Write me a sales strategy

For whom? Over what time frame? With what constraints? For what market? With what appetite for risk?

What should we do about X client who isn't happy with the service?

What's the client's history? What's the service? What's the complaint? What's the company policy? What are we trying to get out of this? What is the client looking to get out of this?

AI will usually try its best. It will produce something fluent, confident and well-structured. But without context, the result is often generic or entirely wrong for the situation.

The issue is not that the AI is bad. Often the issue is that we have given it the equivalent of "Who was that guy?"

Context is not decoration

We sometimes treat context like it is extra detail. Something optional. Something we add if we have time.

I have done this in the past with people! I think about times I've had new developers on projects or juniors working with me and I've given them something to do and just expected they knew how to do it and how to do it the way it needed to be done for the given scenario - without telling them.

I think about a comedy sketch from the great Kevin Hart where he describes asking his kids to do something and them doing it in a way he hadn't imagined having never specifically articulated the way he wanted it doing.

"No, No I want you to do it in the way I imagined you doing it in my head"

But context is not decoration - it's what makes the answer or result possible.

Without context, the answer might still sound intelligent but really, it is just a guess with confidence.

Decision making

Beyond communication, as humans we do not usually make decisions from isolated facts. We draw on a huge amount of grounding.

Most of this happens automatically. We do not consciously list every input before making a decision but those inputs are there.

Our decision making process is built from context and grounding.

At work this becomes even more complex. Think about the things we additionally rely on to make decisions in our jobs:

We use all of that to make informed decisions.

So, if we need all that context and grounding to make an effective decision, why would we expect an AI model or agent to make a useful work decision if it cannot access the same grounding a human would use?

AI that can act

This whole conversation becomes even more critical when we move from AI chat assistants to AI Agents.

If I ask Claude, Chat GPT or Copilot a vague question, I might get a vague or generic answer. Annoying granted, but it's cost me some wasted time and it's recoverable.

But Agents may have been given tools to use, it may have access to data or systems, it may have workflow remit, it may be able to create tickets, update records, send emails, trigger processes, recommend pricing, escalate issues, or interact with customers.

So now the question is not only did the Agent give the right answer, it also becomes did the agent take the right action.

And that changes the risk profile completely.

As agency increases, so does the need for grounding.

Key lesson

This is the key lesson in my opinion. An ungrounded agent is not a digital worker. It is automation with confidence, access and remit to act.

Once AI can act, context is no longer just about answer quality. Context becomes a safety mechanism, a governance mechanism, a control mechanism.

An Agent operating in a business environment needs to understand more than just the immediate instruction. It needs to understand the context of the work, the meaning of the data, the boundaries of its role, what knowledge it can use, what actions it can take and when it needs to involve a human.

Closing thought

So to bring it back to where we started.

If someone asks me "Who was that guy?", I need more context.

If a human colleague is making a decision, they need grounding.

If an AI assistant is generating an answer, it needs context.

And if an AI Agent is going to act on behalf of a person or organisation it needs context, grounding, boundaries and governance.

AI does not replace context, it depends on it. The future of AI at work will not be defined by who writes the cleverest prompt. It will be defined by who connects intelligence to the right context, at the right moment, for the right decision.

Or to put it another way:

You're gonna have to be more specific

#agentic #ai #context #grounding

#agentic #ai #context #grounding