Notes from inside.

Note 01 · August 2026

Every business is getting a brain.

The models are a commodity. The brain you build with them isn’t. Where I reckon the next five years actually goes.

Every year the models get cheaper and better, and every year the stuff built on top of them gets easier to copy. The chatbot someone sold you in 2024 is a free feature today. That curve doesn’t flatten. So if you’re an owner wondering when to “do AI”, the honest answer is that the AI itself is the least interesting part of the decision.

What I reckon is actually happening: every business is slowly getting a brain. One place where everything the business knows lives. Every job, every quote, every client conversation, every hard-won lesson about which supplier actually answers the phone after 4pm. The businesses that start capturing that now will own something nobody can copy, because it’s theirs. The ones that don’t will keep paying the tax they pay today: the owner as the single point of failure, knowledge walking out the door with every resignation, Sunday nights lost to paperwork.

I run my own company on one of these brains. It reads my email, takes calls, reconciles books, briefs me every morning, and remembers everything I’d otherwise lose. That’s my actual Tuesday.

And the thing I keep noticing is that the value was never the automation. The value is that the business now knows itself. Decisions get faster because the information is already assembled. A new starter learns in days what used to take months. The owner can leave for a week and the place still knows what it knows.

The models will keep changing underneath all of this, and it won’t matter. Whoever holds the captured business, holds the business.

Asgard Bishop · Brisbane  ·  back to the map

Note 02 · August 2026

AI is the cheap part.

Every automation you can buy gets undercut within a year. The real cost is the setup, and the return curve is the whole reason to pay it.

The chatbot someone sold you in 2024 is a free feature today. That keeps happening. Any automation you can buy off a shelf gets copied and undercut inside a year, because the models underneath get cheaper every quarter and everyone builds the same wrapper on top. If your AI strategy is buying tools, you’re renting something with a falling price.

The part that costs real money is the setup. I’ll be straight about that, because most people selling this stuff aren’t. Getting AI properly into a business takes weeks of unglamorous work. Someone has to sit inside the operation, watch how a quote actually gets made, find where the hours leak, wire the system in, and keep tuning it until the crew trusts it. There’s no shortcut through that. It’s time and energy, and it’s the honest bill.

But look at what happens after. A machine that can think starts covering its own cost. I reconciled 616 invoices across a £142K expense book for a UK training business, two years of receipts, the kind of job that quietly eats someone’s month. The system did it and then it stayed. It’ll do next year’s too, without being asked twice.

That’s the shape of the whole thing. The cost is a lump you pay once, up front. The return is a curve. Every workflow you capture makes the next one cheaper to build, because the system already knows your pricing and your way of doing things. And when the models improve underneath, which they do every few months, your whole setup gets smarter for free. Nobody sends you an upgrade invoice. The thing you paid to set up just wakes up better.

Compare that to any other asset an owner buys. A ute depreciates the day you drive it off the lot. This runs the other way.

So yes, there’s an upfront investment, and anyone who tells you otherwise is selling you the 2024 chatbot. We price a pilot as one workflow in 30 days for exactly this reason. Small enough to prove the curve exists. After that the maths does the arguing.

Asgard Bishop · Brisbane  ·  back to the map

Note 03 · August 2026

The five steps, in order.

Question it, delete it, simplify it, speed it up, and only then automate. Most AI companies start at step five, which is how broken processes end up automated at speed.

Musk has an algorithm he beats into every engineer at SpaceX, and it works better on small business paperwork than it does on rockets. Five steps, strict order.

First, question every requirement. And a requirement has to arrive with the name of the person who set it and their reason. “Head office wants it” doesn’t count. “We’ve always done it that way” definitely doesn’t. If nobody can name who wanted a step and why, it’s a suspect. Second, delete the step. You’ll delete too much sometimes. That’s how you know you’re deleting enough, add the odd one back later. Third, simplify whatever survived. Fourth, speed it up. Fifth, and only fifth, automate.

Most AI companies start at step five. They walk into a business, see the quoting process, and wrap software around it exactly as it stands. Take a quote I watched get written three times. It went into a notebook on site, then into a spreadsheet at home that night. Then someone typed it a third time into the invoicing tool. Automate that as it stands and congratulations, you’ve built a machine that does the triple entry faster. Run the steps in order and two of those copies just die. Question who ever needed the spreadsheet. Nobody could say. Delete it. Now there’s barely anything left to automate, which is the point.

You can only run these steps from inside the work, though. From the outside, every step looks necessary. Every field on the form looks load-bearing and every approval looks like it’s protecting something. I’m on a landscaping crew in southeast Queensland as a labourer, on the tools, and that’s the only reason I know which parts of the quoting chain are real and which parts are scar tissue from a problem nobody remembers. The voice-to-quote system I built there only exists because the walk-around voice note was already how a job got scoped. We deleted the retyping. What survived, we automated.

An outside consultant would have automated all of it and sent the invoice.

Asgard Bishop · Brisbane  ·  back to the map

Note 04 · August 2026

Your data is the business.

A business that runs on paper trails sells for multiples of one that runs on memory. Most owners are sitting on the second kind and calling it an asset.

I took a cafe on Straddie from $7K a week to $1.8M a year, working inside it the whole time. Five staff became 35. And the thing that scared me at every stage of that climb was how much of the business lived in heads. Mine, mostly. Every time a good person quit, a piece of the operation walked out the gate with them, and I paid in months to grow it back inside someone new.

That’s the standard setup in small business. The owner is the database. Prices, supplier quirks, which customer pays late, the real reason the last hire didn’t work out. All of it in one skull. Go on holiday and the business runs dumber for two weeks. Get hit by a bus and there’s no business at all, there’s a shed full of gear and a phone number.

An intelligent business is built in layers, and you can picture them without a diagram. At the bottom sit the raw records, every invoice, quote, call, and job photo, captured instead of scattered. Above that live the workflows and the decisions, the quote plus the reasoning that priced it. Then a memory that connects the lot, so a question like “what did we charge for the last job like this one” gets answered in seconds instead of an afternoon of digging. On top of the memory sit agents that act on it. They chase the invoice, draft the quote, text the client back, put a brief on your phone at 6am. You steer.

I run my own company this way. Its memory is a graph of over 1,300 connected notes, and it grows every day whether I’m having a good week or not.

The endgame is worth saying out loud. A buyer is paying for confidence the thing keeps earning after you leave. A business that runs on paper trails sells for multiples of one that runs on the owner’s memory, because the first is transferable and the second is you. Capture everything now and you’re building the version that’s sellable. The data is the business.

Asgard Bishop · Brisbane  ·  back to the map