Sometimes I wonder if we’ve stretched the word “AI” so far that it’s lost its shape.

So instead of debating definitions, I find it easier to just… go visit a couple of typical businesses and see what’s actually happening on the ground.

Nothing scientific — just a blue-collar crew and a white-collar professional in Sydney.

Two snapshots.

Two very normal operations.

And we can ask, quietly, without judgement: Is any of this really AI? And if they wanted the real thing, what would the road actually look like?

1. A Tradies Crew in Marrickville — Let’s Start Here

Five guys.

Three vans.

One owner who still answers the phone himself.

The usual mix of blocked drains, hot water systems, and small rewiring jobs.

If you ask him whether he’s using AI, he’ll tell you “Yeah mate, we’re all over it.”

Then he shows you:

  • a scheduling app that sends appointment reminders
  • routing software that finds the fastest path
  • a form that autofills based on past jobs
  • an email tool that makes his quotes sound friendlier

All useful.

All convenient.

But every single one of these things is automation.

Smart-ish rules.

Predictable outputs.

Nothing in the system learns from the last 500 jobs.

Nothing adapts on its own.

Nothing says, “Blocked drains in Marrickville usually run long if Technician B is on duty.”

It’s not a criticism — it’s just reality.

He’s using tools that save time, not tools that think.

If he did want real AI… what would actually have to change?

This is where the conversation gets uncomfortable.

He would need:

  • consistent logs of job types
  • accurate timestamps
  • photos labeled correctly
  • outcomes recorded the same way every time
  • material usage tracked without gaps
  • customer notes that don’t drift into chaos

 

Basically, he’d have to become the kind of person who stores records like a librarian with OCD.

And his team — who already push back on filling out job notes — would have to do the same.

He’d also need:

  • enough jobs to form patterns
  • a clean data warehouse
  • someone (or something) maintaining the model
  • the stomach to keep feeding it data every week

 

And then we hit the obvious question:

For what?

To maybe predict job durations a bit better?

Or suggest van inventory slightly more accurately?

Nice-to-haves.

But not life-changing.

And once a new model comes out in six months — which it will — the whole thing might need to be rebuilt anyway.

So you can see why most tradies teams never get past automation.

Not because they’re behind, but because the effort-to-benefit ratio is quietly upside down.

2. A Small Law Practice in the CBD — Different Industry, Same Pattern

Two partners.

One paralegal.

A modest office with too much paperwork and a very tired espresso machine.

Ask them about AI and they’ll tell you they use it for:

  • summarising documents
  • cleaning up emails
  • drafting templated letters
  • turning messy notes into polished paragraphs

Again — automation.

Language transformation.

Convenient, yes.

Intelligent? Not really.

Nothing looks at their past cases.

Nothing identifies winning arguments.

Nothing spots risky clauses based on their own precedent library.

It’s all surface-level help, not internal learning.

If they wanted real AI… here’s the part that gets thorny

They’d have to:

  • digitise every past case
  • label documents consistently
  • structure their notes
  • store outcomes in a uniform way
  • keep adding new data after every matter
  • maintain privacy and compliance
  • pay for model retraining or fine-tuning

 

It’s basically building a tiny in-house Westlaw.

And again… the payoff is ambiguous.

Would it help?

Yes.

Would it justify months of data cleaning and ongoing maintenance?

Hard to say.

Especially in a world where new models appear every quarter and make last quarter’s fine-tuning obsolete.

The lawyers I know don’t have the energy for that.

They barely have the energy to finish their billing.

And I don’t say that as an insult — just an observation.

Where I Land After Visiting These Two Businesses

When I look at these two cases — one blue collar, one white — the pattern feels pretty clear:

Small businesses think they’re using AI because the tools feel clever.

But under the hood, it’s automation wearing an AI badge.

And if they ever wanted “real AI,” the kind that learns from their history, adapts to their patterns, and makes decisions based on their data…

They’d need:

  • obsessive record-keeping
  • stable workflows
  • enough volume for patterns to matter
  • the patience to keep feeding the machine
  • the budget to maintain it

 

It’s not impossible.

It’s just misaligned with how most small businesses naturally operate.

They’re built on intuition, speed, and experience — not data discipline.

And maybe that’s the quietly honest conclusion:

I don’t think most small businesses are falling behind in AI adoption.

I just see people using simple automation because it fits the way they work. And most days, that is enough.

Real AI would ask them to log everything they do, keep the data clean, pay to maintain the system, and trust that the benefits will eventually show up.

Maybe they would. Maybe they wouldn’t.

Hard to know. Especially when the models keep changing.

So the story feels less about missed potential and more about fit.

Small businesses run on momentum and practicality.

AI still expects a level of order that their days rarely allow. And I keep wondering if the tool should meet them where they are, rather than the other way around.