I use AI a lot
That sentence tends to invite one of two equally unhelpful assumptions.
Either AI is doing all the work for me and I am sitting nearby taking the credit.
Or it is basically fancy autocomplete and therefore not particularly important.
Neither is true.
I use AI because there are now a lot of things I can get done faster, explore more deeply or organise more effectively than I could before.
That matters when you work independently.
But the interesting bit is not "AI can write an email".
If that is the revolution, we may all go home.
The interesting bit is what happens when one person can coordinate research, code, content, analysis and administration without needing a separate human being sitting in every chair.
That changes the shape of a small business.
The useful AI work is often quite boring
Most public demonstrations of AI are designed to look impressive.
Make me a logo.
Write me a viral post.
Build me an app.
Create a cinematic video of a hamster running a hedge fund.
Fine.
I am much more interested in the dull things.
Take a messy set of meeting notes and turn them into a clear list of decisions and unanswered questions.
Compare two versions of a technical document.
Go through a large pile of information and identify contradictions.
Turn a rough voice note into something structured enough that I can actually work with it.
Check whether a piece of work has done what it claimed to do.
Prepare the first pass of a repetitive admin task so I can spend my attention on the part that requires judgement.
None of that makes a particularly exciting demo.
It does make the day substantially less irritating.
I am not paid for pressing the buttons
This is the distinction I think matters most.
If the value of what I do is simply that I know which menu to click in a piece of software, then AI is a fairly serious problem for me.
And, frankly, fair enough.
Crack on then. Go for it.
The value should be somewhere else.
What are we actually trying to achieve?
Is this the right problem?
What stack makes sense?
What has the client forgotten to mention?
Which option creates a maintenance nightmare six months from now?
What does "finished" actually mean?
Where is the evidence that the thing works?
Those are judgement questions.
AI can contribute to them. It can challenge them. It can find information I missed.
But somebody still has to own the decision.
First drafts are cheap now. Good decisions are not.
This is one of the biggest changes.
The blank page has lost a lot of its power.
If I need the structure of a proposal, documentation, a technical plan or an article, I can get to something workable very quickly.
That is useful.
It also means the first draft is becoming less valuable.
The value moves to the edit.
What is wrong?
What is missing?
What sounds like nonsense?
What is technically unsafe?
What has the model confidently invented?
What should be deleted because nobody needs to read it?
The ability to produce more words is not necessarily an advantage.
Sometimes the advantage is being able to remove 40% of them without losing anything useful.
Context is the real leverage
AI becomes much more useful when it knows enough about the job to stop guessing.
That does not mean throwing every private document you own into the nearest chatbot.
Data handling matters.
Permissions matter.
The system you use matters.
But within appropriate boundaries, good context changes everything.
A model that understands the project brief, the previous decisions, the technical constraints, the client’s language and the work already completed is far more useful than repeatedly starting with:
Hello. Please act as an expert.
If I have to explain the entire world from scratch every time, I have not built a workflow.
I have acquired a very articulate intern with amnesia.
Orchestration is more interesting than prompting
People talk a lot about prompt engineering.
Prompts matter.
But I find orchestration more useful.
Which tool should do which part?
What information does it need?
What should be checked independently?
Where does a human decision sit?
What evidence proves completion?
When should the process stop?
That is much closer to managing a small team than using a magic text box.
One agent can produce something.
Another can criticise it.
A third can verify facts.
You can keep source material in a place they can all refer to.
You can make the output land in the system where the next stage actually happens.
Then the human stays responsible for the shape of the whole thing.
That is where AI starts becoming operational rather than entertaining.
Do not automate the human bit out of the work
There is an obvious danger here.
Once you discover that something can be automated, you start wanting to automate everything.
I do not think that is the goal.
Some client conversations need a person.
Some disagreements need judgement.
Some creative decisions depend on taste.
Sometimes somebody says one thing but quite obviously means another, and there is no five-step framework for knowing that. You just know because you are dealing with another human being.
That matters.
The aim is not to build a business in which nobody has to think.
The aim is to stop spending human attention on tasks that do not deserve it.
A small business can now have a much bigger operating surface
That is the part I find genuinely interesting.
A solo operator can research more.
Test more.
Document more.
Produce more formats.
Keep better records.
Automate repetitive follow-up.
Build things that previously would have required several different specialists.
Not perfectly.
Not automatically.
And certainly not without knowing enough to spot when the machine has wandered off into the bushes.
But the leverage is real.
The question is what you do with it.
You can use AI to fill the internet with more generic material.
Or you can use it to remove enough administrative friction that you have more time for the work only you can do.
I know which one interests me.