I have held many roles across my career but two have shaped how I see almost everything.
One is operational risk management and risk advisory, years spent inside institutions built on the assumption that things can and will go wrong and you better be prepared for that.
The other is something harder to put on a CV. I have always been deeply entrepreneurial, with side projects running parallel to whatever my official title was, taking risks, learning from failures, going after things that were untested and often very much off script.
That combination gives me a particular lens. I am drawn to new territory but I also never stop asking what could break, which is exactly why I find myself with mixed feelings about letting AI run my life and business.
Not against it, but not willing to go in blind either.
The space in between, how to use AI as a builder rather than a regulator, feels surprisingly empty. You almost have to choose a side. Either you are deep in frameworks and risk taxonomies or you are automating everything in sight at any cost. Both have their place but there is almost no conversation happening in the middle, where most of us actually live: trying to use the tools, move fast, and still understand what we are getting into.
So that is what this is.
Automate everything. Or should you?
I definitely feel the pressure, from myself first and foremost, to keep up with the AI transformation. I want to be as efficient as my most AI-fluent peers, which means learning fast and handing off the predictable, repetitive parts of my work.
The speed of doing business today is relentless regardless of your industry. Either you have a team behind you, and even they are catching up, or you are pretty much a one woman show, which means staying relevant requires automating and agentising as much as possible.
In theory that sounds amazing. Never having to properly clear your inbox, write things from scratch, or manually keep track of every commitment, obligation, order, and reservation. The ultimate personal assistant, if only they could also do the dishes.
Then you go online to figure out how and suddenly everyone is selling you one prompt miracle. Influencers and gurus from every corner of the internet telling you to comment something on their post so they can send you their golden freedom ticket prompts to automate and outsource pretty much all your life, business thinking, and project management to AI.
It just sounds a little too good.
Or maybe I am just too risk-trained. Either way I find myself one foot in and one foot out, and here is what I have actually figured out.
What you are actually doing when you automate
Before getting into what to automate and what not to, it helps to be honest about what is actually happening when you hand something to an AI tool (I mean agentic usage like Claude co-work):
You are delegating judgment.
Every time you let a tool draft your emails, prioritise your tasks, or make decisions on your behalf, you are outsourcing a piece of thinking that used to be yours. That is not inherently bad. It frees up capacity for the things that actually need you, but it is worth being conscious of, because judgment is a skill and skills fade when you stop using them.
You are also creating a dependency.
The more deeply a tool is embedded in how you work, the harder it becomes to function without it. That is fine as long as the tool keeps working the way you expect. It becomes a problem when it changes, disappears, or starts behaving differently without you noticing.
And that last part is more common than people realise. AI models get updated quietly. The tool you evaluated and decided was good enough six months ago is not necessarily the same one you are using today.
None of this is a reason not to automate. It is just useful context for the decisions you are about to make.
The questions worth asking before connecting anything
So when I decide what gets access autonomously, I run a little check list. Not a list of things to be scared of, but just a simple framework for making the decision more consciously.
What is actually in here? Creative files, project assets, and templates are low risk. Client communications, relationship history, and anything involving other people’s information are higher stakes, not because the tool will necessarily misuse it but because you are making a decision about someone else’s data without their input. (The fact that most cloud services already process your data does not change this, it just means the question is how many additional layers you are consciously choosing to add).
What happens if this tool disappears or changes tomorrow? Can you still function? Is your work exportable? Have you built something that works independently of the tool or something that only works because of it?
Am I making this decision consciously or just clicking next? Most people accept terms without reading them. That is fine for low stakes tools. For anything touching your professional relationships or the information infrastructure of your business, it is worth a minute of actual thought.
Would I be comfortable if my clients or colleagues knew this tool had access to this? That question tends to cut through the noise quite quickly.
So where does that leave us?
I really think that you and me should be using these tools.
The efficiency gains are real and the pressure to keep up is not going away. After all, I don’t think we have anymore the luxury of asking whether we should but rather how can we do it consciously.
So the whole point becomes actually understanding what are you doing when you delegate to a tool, make those decisions deliberately, and stay sharp enough to catch it when something is not working the way it should.
That is the bridge worth building. Not between automation and abstaining, but between moving fast and knowing what you are moving into.