AI isn’t coming for my job. Yet.

In August, I wrote about what happened when I tried to make about fifteen years of communications experience explicit enough for AI to use.

Or, to put it rather less grandly, I tried to help a machine understand why “we need an email” is not always the beginning of a communications strategy.

I wrote about that experiment here.

There was also a slightly more selfish motive.

If AI was eventually going to replace parts of my job, I thought I might as well try to build it myself. Ideally before someone else did and charged me a subscription to use it.

The interesting thing was that some of the thinking was easier to codify than I expected. Start with the outcome. Separate the business problem from the communications problem. Don’t confuse activity with impact. Don’t build a twelve-channel campaign when three things will do.

Fine.

The harder bits were the things I had stopped noticing I was doing.

Why does this brief feel wrong?

Why is now a terrible time to say this?

Why do I believe this technically accurate sentence is going to land like a brick?

And why am I fairly sure that what everyone is calling a communications problem is actually a process problem wearing a lanyard?

Towards the end of my time at Amazon, one version of this idea became something I called D.A.V.E.—either an early AI communications reasoning agent or a slightly disturbing attempt to create a digital version of myself.

The name came first. I worked backwards to invent the acronym, which is probably how quite a lot of corporate acronyms come into existence.

The idea was that some of the senior stakeholders I worked closely with could ask D.A.V.E. the sorts of questions they might otherwise have asked me. I connected it to relevant material and tried to make explicit some of the experience and judgement I’d accumulated over the years.

The final D.A.V.E. wasn’t anywhere near as capable as me.

This was reassuring on several levels.

But it was useful.

More recently, I’ve pushed the idea much further. What was once a long and complex instruction set has become a Copilot agent, and last week I demonstrated the latest version to a group of senior communications leaders.

It can already help diagnose a communications challenge, identify the intended outcome and build a response proportionate to the problem.

I’m now trying to teach it something harder: is this actually a communications problem at all?

Because the more I develop the agent, the clearer it becomes that professional judgement does not sit neatly inside a framework.

It needs context.

An agent can know your strategy. It can read previous communications. It can understand your audiences, your channels and your tone of voice.

But to make the kind of judgement an experienced professional makes, ideally it would know much more than that.

What else is happening in the organisation?

What have people already been asked to absorb?

Where is trust strong, and where is it fragile?

The same announcement can be sensible in one organisation and completely wrong in another. It can even be right in March and wrong in September.

Context changes the answer.

Leaders introducing AI into organisations have the same problem. Giving people access to the technology is relatively easy. Deciding where it should make decisions, where people need more context and where human judgement still matters is harder.

This is one reason workplace communication is such an interesting place to study AI. The quality of the answer depends on much more than the information available to the model. It depends on what is happening around the work.

I’ve written before about how agents need more than instructions if they are going to become genuinely useful participants in work.

An AI might have read every communications plan your organisation has ever produced and still miss the thing that matters most.

That everybody is exhausted.

That three reorganisations have landed in six months.

That one more reference to an “exciting transformation” may require medical intervention.

That is the gap where professional judgement starts to matter.

My current working definition is pretty simple.

Professional judgement is what we use when there isn’t one obviously correct answer.

It is the ability to make a reasoned call by combining professional knowledge, experience and context. The “context” bit increasingly feels crucial. So does the fact that I’ve called this a working definition. I’m not convinced it is the whole answer.

I’ve also argued before that AI can increasingly produce the line, while humans still have to make the call.

I’m just much less satisfied now with saying “judgement” and assuming we all know what that means.

Because there is another problem.

If AI starts doing more of the work through which people traditionally acquired experience, what happens to the development of judgement itself?

A relatively new communicator can now ask AI to produce a pretty sophisticated plan. That could be enormously helpful.

But does producing better work automatically make you better at judging the work?

I’m not sure it does.

Some judgement is built through the awkward stuff. The bad first draft. The idea that sounded clever until somebody asked one difficult question. The campaign that was technically fine but landed badly. The stakeholder conversation that made you realise you had misunderstood the problem.

That is the professional scar tissue.

And if AI helps us skip some of those mistakes, which is obviously useful, we may also need to think much harder about what else was being learned through making them.

For me, this is becoming one of the more interesting questions about the future of work.

What happens when AI gives people some of the outputs of experience before they have had the experience themselves?

We’ve always borrowed expertise from colleagues, specialists and mentors. What feels different with AI is how easily we can get the answer experience might produce without the conversation, challenge and feedback through which we develop judgement of our own.

And it tends to be very confident about it.

AI is making knowledge easier to access, but that doesn’t automatically mean people or organisations are becoming better at recognising what matters.

So rather than keep theorising about it until I bore both of us, I want to understand what this actually looks like in practice.

That is why I’m pleased to be collaborating with Definition and the Northern Comms Collective on a new piece of research into how generative AI is affecting communications.

You can read more about the wider research programme here.

We are approaching it from two different directions.

My part goes deeper into the judgement question—and it is deliberately broader than internal communications.

If you work in communications, whether that is internal comms, corporate affairs, PR, media relations, public affairs, change, employee communications or something adjacent, I’m interested in your experience.

The study works more like a research conversation. You can speak or type, pause and take time to think. Your answer doesn’t need to be polished.

It uses Given Time, the qualitative research approach I’ve been developing around a simple principle: if people give us their time, the experience should feel worth giving it.

Alongside this, Definition and the Northern Comms Collective are running a short survey focused specifically on internal communications.

It is designed to build a broader “state of the nation” view of how AI is being used, where it is creating value, what is holding adoption back, and how people feel about confidence, trust, governance and professional judgement.

The two approaches should give us different things.

The NCC and Definition survey should give us breadth across internal communications. The research conversations should give us much greater depth into some of the thinking underneath communications work.

Then we can put the two together and see what emerges.

I genuinely don’t know what that will be.

I have hypotheses: that some of the most valuable things experienced professionals do have become almost invisible, and that organisations may need to become more deliberate about how judgement develops as AI changes the work.

But evidence that tells me something different will be considerably more interesting than proving myself right.

For now, I keep coming back to one thing.

AI is making more of the work easier to produce. That may make the thinking underneath the work more important, not less.

And if we are going to keep saying that human judgement is one of our advantages in an AI-enabled world, it feels sensible to understand what it actually is, where it comes from and how we avoid automating away the experiences that help us develop it.

If the question interests you but you don’t want to take part, you can also register to receive the findings when the research is complete.