Twelve Months of AI, and Losing Sleep

Domain expertise. The knowledge of your industry. How your business runs. What the regulations require. What the operational rhythm of a season looks like.

Human adoption. How people work in practice. The trust your team needs in a tool before they will rely on it. The accountability your customers expect. The policy that gets followed, not the one that sits in a folder nobody opens.

Technology. The models, the platforms, the agent orchestration. The audit and governance layer.

When all three come together, and only when all three come together, AI adoption succeeds.

The chasm is widening.

The people I talk to in the AI community are losing sleep over this. They are running at a level of detail and pace that the leaders in maritime organisations I work with have little time to track. The gap between those two worlds is widening. You can do everything you can to learn AI on your own. But, and I say this as gently as I can, you will not get there with the time and focus available, unless you do one of three things: free up more of your own time to focus on this; nominate someone in your organisation and free up their time to focus on this; or work with someone external, like me, who can filter what you need to know and turn it into something you can put into practice.

How your business is found has changed.

AI chat windows are answering questions on behalf of users instead of returning ten blue links like the days of traditional SEO. Whether your business is discovered depends on whether the AI's picture of your entity is accurate, complete, and consistent across every platform. I had to do this work for myself. There is another Kristina Agustin. She is a California realtor. Until I strategically addressed this, Google AI could not tell which me was connected with Southern Sky AI and which was a realtor. I was not impressed and took this on as a personal mission to rectify, to my family's entertainment. Being findable in 2026 is making sure the AI knows exactly who you are, what you do, and which Kristina is the right one.

Context matters more than prompting now.

A year ago we were learning to write good prompts. Today the models work out what we mean. What matters now is the context. Your documents, your data, your written instructions, locked and loaded so they inject into everything you do. This is not about uploading documents to chat, or how your model starts to know your preferences over time. Context engineering is deliberate. You decide what the model sees, when, and in what order, and you design your systems around it.

AI is already being used invisibly in almost every organisation I have met with.

And it is more than just data risk. A recent conversation with a maritime CEO described it as a habit risk. People are getting stuck in habits with certain tools that may not be ideal for their entire operation, and using them in ways leadership does not understand. By the time leadership decides to implement a strategy around certain tools and systems, the behaviour has hardened and change management becomes more challenging. McKinsey found that employees are using AI three times more than their leaders think they are.

People have a limited understanding of what AI governance is.

Which is fair enough, because I did not understand it either to start with, and I am a lawyer. The term is confusing. AI governance is the whole system you put around AI in your business. It includes a policy, the rules about who can use AI, what they can put into it, what it is allowed to do with company data. But it is more than just the policy. It is also what gets tracked, what gets logged, whether decisions made with AI can be reviewed six months from now, whether your team knows AI was used, whether your clients know, who is accountable when something goes wrong, and how often the rules get revisited as the technology changes. A policy is the speed limit sign. Governance is the sign plus the camera, plus the licence points, plus the review of road safety every five years. The sign on its own does nothing if nothing is tracking it. Most people in our industry are deeply aware of the sensitivity of owner and client data. Most have not known how to engage with AI without accidentally leaking private information. So they have either not used AI at all, or used it on best judgment. Best judgment is not a policy. It is not defensible, trackable, or repeatable.

People still think AI is the chat window.

It is not. The best analogy I came up with is a food truck. The kitchen inside the food truck is the intelligence. That is the brain. The AI. The question is how you get to it.

People are moving between models. Are we waiting for a clear leader before we commit?

I do not think the answer is what model wins, because the answer changes almost by the day. The way to handle it is to stop relying on the model to remember what it knows about you. Build a DNA file. Everything the model should know about you, your business, what you are trying to achieve. Point the model at that file every time. That is what context engineering is. Stay model-agnostic as much as you can. If you are working on more complex problems, break the work up into projects, with a folder per project, and fill each folder with the relevant data and knowledge. You will notice the answers get much more accurate and specific.

What happens when one person writes the agent code the rest of the organisation comes to rely on, and only they know how it was built?

The first step is always have redundancy in any organisation. Easy to say, not so easy to do, because there is not yet an abundance of people who understand how this all goes together. You can back the code up to a repository like GitHub. And what I like to encourage, and see as an important part of working with businesses going forward, is to have what I call an AI champion in-house. It might not be one person. It might be two or three. Having those people learning and coming along for the ride is the answer.

How do we get leadership on board when leadership does not have the time?

The three options I gave earlier are the honest answer. Free up your own time. Nominate someone and free up theirs. Or work with someone external. There is no fourth option that works. This is the thing I have found hardest about the last year. I am running at one pace, in one community, where this is normal. The leaders I most want to help are running at a different pace, in a different community, where there is not enough time. Sometimes I feel like Chicken Little. Sometimes I am not sure if the right move is to wait until something forces the conversation, or to keep saying it gently until the room is ready. The $11.5 billion the labs spent in May suggests I am not the only one trying to work that out.

If an AI agent writes a CV and another AI agent is filtering CVs, are we just getting AI talking to AI?

Yes, and you have to engineer against it. The model is not intelligent. It is a mathematical machine looking at the highest probability of an outcome based on what it has been trained on. Past data carries past bias. If 98.5% of people successful in a role up to this point were men, the model will infer that men have a higher rate of success and weight accordingly. You have to be aware of it and engineer against it. Force a 50-50 split on the data set. Apply positive constraints. Where AI is making decisions that affect a person's rights or interests, you are about to have a legal obligation in Australia to disclose what data the system is using and what kinds of decisions it is making. From 10 December this year. The speed limit sign has been there for years. From December, the camera goes live.

Are you aware of the legislation coming in December?

If your turnover is above three million dollars in Australia, you are an APP entity. From 10 December this year, you must disclose what personal information your AI systems use, the kinds of decisions they make, and where those decisions could significantly affect someone's rights or interests. The obligation captures systems you already have running today. This is policy work and mapping work. Understanding what your AI is touching, what decisions it is making, and stepping back to decide whether you want it touching those decisions at all. Hiring is the example I expect to see real pressure on first.