AI Era is Happening
You may have heard about a new powerful AI tool coming online in recent weeks — or even in recent days. You may feel the need to learn these new tools as quickly as possible; otherwise, you could be left behind.
And with that comes another fear: losing your job, the value of your skills, your experience, your knowledge, your capabilities that took years to build — or even your sense of purpose.
That's AI anxiety. And if this sounds like you, you are not alone. I feel it too.
I still remember that I wasn't too worried about AI one year ago. I used ChatGPT from time to time, found it useful, but I did not see generative AI as a serious threat to my career.
A few months ago, I started hearing more people talk about Claude and how powerful it was. At first, I thought it was just another AI tool. However, it wasn't until recently that I spent more time researching it and fully realized what AI is already capable of.
AI agents are indeed game changers.
They are no longer just smart chatbots. They can already perform many tasks typically done in "laptop jobs." For example, they can automate work routines and workflows, coordinate across different tools and systems, retain memory, and even fix errors by themselves.
On one hand, this could allow white-collar workers to spend less time on administrative tasks and focus more on higher-value work. On the other hand, it may also mean that people will need to learn how to work with AI in order to keep their place in the office. Some jobs may see reduced demand — or even disappear entirely.
At the same time, we are seeing more and more headlines about AI-related layoffs from large companies such as Meta.
No wonder so many of us feel anxious or even scared about the future.
To better understand what AI agents could mean for me personally, I decided to break down what I do as a risk manager in large infrastructure projects (especially offshore wind) and compare those tasks against the current capabilities of AI, at least based on my own understanding.
Breaking It Down: Four Levels of AI Capability
Rather than asking a simple "can AI do this or not," I found it more useful to think in four levels:
AI OK — AI can do the work. You still need someone to check the output, but the heavy lifting is handled.
AI Assisted — AI can help significantly, but it's not reliable enough to work alone. A human needs to stay involved throughout, not just rubber-stamp the result at the end.
AI Difficult — AI can contribute something, but the task is complex enough that it's unlikely to do it well without substantial human guidance.
AI Impossible — AI simply cannot do this. Not today, and probably not for a long time.
What This Looks Like in Practice
Let me walk through the actual workflow of a risk manager on a large infrastructure project.
Inputs: where the work begins
A lot of what feeds into risk management is information processing — scanning documents, conducting risk interviews, cross-referencing other projects. This is AI OK territory. AI is genuinely good at reading, extracting, summarizing large volumes of text, and it can generate minutes of meeting (MoM) from risk meetings and summarize updates required for risk and/or mitigations in the system. A reviewer still needs to check the output, but the time savings are real.
External information — market news, supplier signals, regulatory changes — is more nuanced. AI can ingest and synthesize this at scale, but figuring out what actually matters for your specific project requires contextual judgment. I'd call this AI Assisted: useful, but a decision-maker still needs to interpret what it means.
Then there's personal experience and the conversations that happen outside formal meetings — a coffee chat where a colleague flags a concern, a hallway conversation that changes your read on a risk. This is AI Impossible, and I think it will stay that way. These moments depend on relationships, trust, and the kind of intuition that takes years to develop. No model has access to any of that.
Functions: the core of the work
The risk and mitigation workflow — identifying risks, assisting assigning ratings, tracking mitigations — is AI OK for much of the process. AI can process inputs, suggest categorizations, and help maintain structure. But approving a mitigation strategy is a different matter. That's a decision with real consequences, and it needs a decision-maker, not just a reviewer.
The risk register itself sits in an interesting place. AI can help maintain and update it, but the entries reflect judgment calls about likelihood, impact, and ownership. I'd rate this AI Assisted — and I'd want a decision-maker involved, not just someone checking for formatting errors.
Outputs: what we actually deliver
The risk report and dashboard — synthesizing data into a structured document for stakeholders is AI OK. It can draft reports faster than a risk manager and often with less effort. A reviewer still needs to check the numbers and the narrative, but the time required to build these from scratch every time is materially reduced.
Quantitative risk analysis (QRA) — running Monte Carlo simulation with P10 to P90 distributions— is AI Assisted. The models can run in risk tool(s) with API, but the assumptions behind those modeling are where the real risk lives, and those need to be owned by someone who really understands the project. Handing that to AI without scrutiny is where things go wrong.
Advanced analysis — mitigation trade-offs, scenario analysis, stress testing — is AI Difficult. These require integrating quantitative outputs with qualitative judgment, stakeholder dynamics, and an understanding of what the organization can actually tolerate. AI can support the modeling, but I wouldn't trust it to own the assumptions and conclusions.
And then there's risk culture — awareness and trust, open communication, accountability and ownership. This is AI Impossible, and I think that's important to say clearly. Culture is a property of teams and people, not a feature you can generate. AI might help communicate risk awareness, but it cannot create the environment where people actually speak up.
The Question I Wasn't Asking
When I started this exercise, I was asking: can AI do my job?
But the more useful question turned out to be: when AI is involved, who is still responsible?
I mapped three human roles across every part of the workflow:
Reviewer — checks the AI output before it's used. Appropriate when the task is well within AI capability and the cost of an undetected error is manageable.
Decision-maker — owns the judgment and its consequences. Required whenever the output informs a significant decision, regardless of how capable the AI is.
None needed — theoretically possible for fully automated internal steps, but rare in risk management.
This distinction matters more than I initially expected. Three tasks can both be rated AI OK, but two of them still need a reviewer and one even demands a decision-maker. This means though AI can do majority of some tasks, a human role is still required to control/intervene at some points.
So, Should I Be Worried?
Yes and no.
The parts of my work that are AI OK — report generation, document scanning, risk register maintenance — will change significantly. If I'm spending most of my time on those tasks, I need to adapt.
But the parts that are AI Assisted, AI Difficult, or AI Impossible — the judgment calls, the stakeholder conversations, the experience-backed instincts, the culture work — those remain firmly human. And in large infrastructure projects, those are often the parts that matter most.
What I take from this exercise is not reassurance. It's clarity. AI is not going to replace the risk manager. But it will change what a risk manager spends their time on — and that means the value of this role will shift accordingly. The anxiety doesn't go away. But having a clear map of where I stand makes it a lot easier to navigate.
Comments are closed.