Hey friends,

Phaedon Stough said something to me this week that stopped me mid-question.

He was explaining how Innovation Bay screens ag-tech startups. When a founder pitches something built for farmers, the first thing they ask is simple. Will this grow the farmer's revenue, or will it save the farmer money? If the founder says revenue, they pass. Not because the claim is false. Because the farmer won't buy it. The farmer already believes he knows how to grow his own revenue line, and he is sceptical of anyone who turns up claiming otherwise. He will swap something out when it saves him on the bottom line. That is it.

Then Phaedon drew the line straight across to everyone else. He assumed the same thing held for traditional businesses. They will engage AI if it saves them money.

I told him he was right, because he is. Every conversation I have confirms it.

But twenty minutes earlier in that same conversation, Phaedon had told me something that pulls in the opposite direction.

His view on where the real value in AI sits is not cost at all. He said the people using AI most effectively are increasing the capability of the business and driving top line revenue without increasing cost. The cost increase is marginal against the revenue increase. That is the prize. Not the saving.

I asked him to confirm it, because it cuts against almost everything you read. Are you saying the saving is real but the bigger prize is the ability to increase revenue several times over? Yes, he said. That is exactly what I'm saying.

So hold both of those in your head at once.

The thing that gets AI bought is cost. The thing that makes AI valuable is growth. Those are not the same thing, and most leaders never make the second move. They buy on the first argument, get the saving, book the win, and stop. The pilot pays for itself, everyone nods, and the actual opportunity sits untouched for another two years.

That gap is the whole story.

I understand why cost wins the argument. Running a business hurts in specific, daily, unglamorous ways, and anything that stops one of those pains gets attention immediately.

We have a name for those pains internally. ROBOTS. Repeated, obligatory, boring, operational tasks. When we sit with a CEO or a board, the big strategic AI conversation often lands with a polite nod. Then we say something small, like this weekly report can write itself, or you can have a briefing waiting for you every morning, and the room changes.

I was working with a construction company recently. Big conversation, big ideas, and I could feel it not quite landing with the CEO. He is always on the road, so I mentioned almost in passing that he could transcribe every call he takes from his phone, push those transcripts into his own AI brain, and stop trying to remember what he agreed to and with whom.

Out of a whole hour, that was the thing that lit him up.

I've stopped being frustrated by that. It is the door. You get invited in through the boring problem, and then you get to have the bigger conversation.

Phaedon was specific about where growth shows up, which I appreciated, because this is usually where people go vague.

Sales was his first example, and not in the way LinkedIn talks about sales. He is not describing list building and cold outreach tools. He means the whole pipeline. When is the right moment to send that email. What should you know before a call, and what should be captured after it. How do conversation flows get handled, how does data get collected without someone typing it in. Put AI through every step of a process that already exists and the salespeople you have become considerably more capable. So you stop needing to hire more of them to grow.

His second example was better, and less discussed. The capability and the decision speed of the CEO. In a startup, the founder is usually the bottleneck. One person trying to make thirty decisions. When that person starts receiving distilled, current information instead of raw noise, the whole company moves faster.

Then he told me something I did not expect. Innovation Bay runs an aggregated job board that scrapes all their member companies' roles into one place. He asked me whether I thought the volume of jobs was growing or shrinking.

I guessed shrinking. It's growing.

Engineering and technical hiring has always led that board and still does, but product is up there, sales, customer support, right across the board. So the growth from AI is being backed by real hires at the same time as agents get deployed. Six months ago the story was that AI would remove customer support and accounting roles. What they are actually seeing is augmentation, not replacement.

Neither of us thinks that is a permanent state. But it is the state right now, and it is worth knowing that the honest data from 200-odd fast-growing Australian companies does not match the headline.

So here is where we land, and it is the same place Phaedon landed from a different direction.

When we walk into an organisation, revenue growth is the metric we are looking for. Not activity. Not the number of tools deployed, not how many people have logins, not how impressive the demo looked in the boardroom.

I've watched a lot of very cool AI demos over the last two years. Cool is not a business case.

There are only two questions that matter, and every AI decision in your business should be able to answer at least one of them without a run-up.

How is this going to increase my revenue?

How is this going to decrease my costs?

If the answer to both is a shrug, or a paragraph about capability and readiness, you have bought a science project. It might still be worth doing. Be honest that that is what it is.

And if the only answer you ever accept is the second one, you are the farmer. You will get your saving. You will not get the thing that actually changes the shape of your business.

The Dumb Monkey move here is not scepticism. Scepticism is fine. It is being permanently impressed.

I see leaders running from one model to another, one tool to the next, genuinely trying to stay on top of a market that changes weekly. Phaedon admitted he has never experienced a market with this much change, and that right now you have to both test and review constantly. He is right about that.

But there's a trap in it. By trying so hard to keep up with what is happening, some businesses miss the bigger thing entirely. They collect tools. Nothing compounds.

My advice to organisations with low AI maturity has not changed. Do not buy a new tool every time you find a new problem. Every tool you add is another vendor, another data path, another governance question, another thing that breaks. Start with a base stack, a good assistant and something like n8n to wire it together, and build on top of that. You will move faster with less risk and you will actually own what you build.

Phaedon's read on where this goes next supports that. He expects the small, hyper-niche AI applications to consolidate into suites, and thinks that within six to twelve months you won't need to commission someone to build you an agent. You will go to a platform and say you need an HR agent, this is what it does, manage it for me. Nobody has won that race yet. It changes by the day.

There is a version of this same gap sitting inside AI governance, and it is going to land on your desk sooner than you think.

Enterprise Monkey has just come through ISO 42001, the international management standard for AI, which as best I can tell puts us among very few organisations in Australia holding it.

Now let me tell you what it does and does not mean, which is not the way most people will sell it to you.

ISO 42001 asks whether your organisation has processes in place. Governance, risk management, robustness, fairness and bias, accuracy. It asks whether things were tested. It does not ask how. It does not technically assure any specific AI system you have deployed.

So the paperwork layer now exists. The proof layer, mostly, does not.

That distinction matters if you have built anything real. If you are running a customer support agent on top of Claude or GPT, you cannot point at the model provider's assurances and consider yourself covered. Your deployment is yours. Your retrieval pipeline, your prompts, your data, your failure modes, your edge cases. Nobody has tested those but you.

Which is the same shape as the revenue problem, if you look at it sideways. The certificate is the thing that gets bought. The proof is the thing that makes it worth anything.

We went and got the certificate because government and enterprise clients need it, and because processes genuinely improve when you are forced to write them down. But the reason I care is the second part. It gave us a discipline for proving our own systems do what we say they do. That is the part I would push you towards, whether or not you ever get certified.

Two examples from the conversation, one his and one mine.

Innovation Bay runs an AI forum for their members, and they are trying to move it to weekly because the pace demands it. Around fifty people were on the last session. One of their founders demonstrated an AI coach he had built himself. He gave it his ten favourite management books, told it the kind of leader he wants to be with his team, and hooked it into Granola so it has every call he takes. At the end of each day it hands him a short read on what he did well and three things to work on tomorrow. Continuous coaching, built by one person, for himself.

Mine is less elegant. We built something we call Agent OS, and it is genuinely just folders of markdown files. Each agent has its own folder and its own task list, also markdown. A scheduled task in Claude Cowork opens the folder, picks up the next task assigned to it, and does it. We had been going through the API and the cost of that processing climbed fast, so we stopped. Now the agents hand tasks to each other through files, and it works.

What that changed is the honest bit. We manage marketing for around eight brands, and we only ever had the appetite to do proper SEO on two of them. Now all eight are in. The agents find the issues, find the opportunities, go into the site and make the changes. Rankings have started moving since we deployed it a couple of months ago, and the citations, the times an AI assistant recommends the business, are climbing too.

We had one full-time SEO person and one content person. We still have one person. That person now covers eight brands.

And it was built by non-technical people in our team. That is the part that would have been impossible two years ago.

Take the last three AI things your business bought or built. Write two numbers next to each one. What it added to revenue. What it took off cost.

If you cannot fill in either number for one of them, you have found the problem. It is not the tool.

Then pick your single biggest revenue-generating process, whatever it is, and ask what it would take to double the output of that process with the people you already have. That question tends to produce a very different list than the one you get from asking what you could automate.

I asked Phaedon a selfish question near the end. How does he sleep. I had gone to bed at two the night before and then dreamt about agents, which I am aware is not a flex.

He said he is sleeping better than he was, and the reason was good. His community has been through real fear. B2B SaaS founders sitting in rooms asking out loud whether they will have a business in twelve months. They are coming out the other side of that into something else, a sense that they can pivot, that they have knowledge and IP worth something, that they can put AI into their own business and grow. He sleeps better because he is hopeful rather than fearful.

I think that is the right sequence, and I think most Australian businesses are somewhere in the middle of it. Fear, then excitement, then, I hope, something steadier. Excitement with governance around it. Excitement that can survive a Monday.

His parting advice was to learn through other people. Connect with communities, find people doing the same thing, reduce the apprehension by talking to someone who is two steps ahead of you. Because if enough of us engage with this properly, we get to shape where it goes, rather than having two or three companies shape it for us.

His actual words were stand up and jump into it. Their membership runs to about 220 people across every state and territory, plus a few in Singapore, New Zealand and the US, roughly ninety of them VCs, and you can apply through the Innovation Bay website.

Just make sure that when you jump, you know which question you are answering. Revenue or cost. If you cannot say, you are not ready to spend the money yet.

I'm giving the keynote at UnLtd's Big Chat on Good AI in Melbourne next Wednesday.

UnLtd is a social purpose organisation that pulls the media, marketing and tech industries in behind its 25 charity partners. Their CMO Nina Nyman got in touch after we took Agents for Humanity to the UN AI for Good Global Summit in Geneva in July, and asked whether the idea could carry a whole session.

My twenty minutes is on using AI against some of the world's biggest problems. A case study on Agents for Humanity, and what I actually saw and heard in Geneva rather than the press release version.

📆 Wednesday 26th August 2026
⏰ 12:00pm to 2:00pm
📍 News Corp, 40 City Rd, Southbank
🎟️ Melbourne Tickets: https://events.humanitix.com/big-chat-good-ai-melbourne

Mark Byrne from Headspring.ai, Candice Ong from Today and Noah Yang from We Are Mobilise are speaking as well, and we all come back on stage together for a joint Q&A to close. Lunch is courtesy of Meta. It is close to capacity, so book if you want a seat.

One ask. If you work in a not-for-profit and you are sitting on a problem nobody has the budget to solve, bring it with you and hand it to me on the day. That is the part I am most looking forward to.

Geelong Young Professionals are launching a Boardroom Series and asked me to go first.

Twelve seats. No slides, no presentation, ninety minutes of open Q&A. The format is the point. In a keynote nobody has to admit anything, you nod along and photograph a slide and go back to work. In a room of twelve, somebody eventually says the real thing out loud. We bought the tool nine months ago and nobody uses it.

That is a different conversation from the one you get at a conference, and it is the more useful one.

📆 Monday 31 August 2026
⏰ 10:30am to 12:00pm
📍 You Yang Boardroom, Level 2, Crowne Plaza Geelong, Little Smyth Street
💵 $25 plus booking fee, light catering included, RSVP by Friday 29 August
🎟️ Book here: https://www.stickytickets.com.au/ut6eqp/board_room_series__aamir_qutub.aspx

Bring the question you would be slightly embarrassed to ask in a bigger room. Those are almost always the good ones.

See you next week,

— Aamir

🎧 Listen to the Podcast Episode on: Spotify | Apple Podcasts | YouTube

📘 Book: The CEO Who Mocked AI (Until It Made Him Millions) by Aamir Qutub

🚀 Dumb Monkey AI Academy

📱 Dumb Monkey AI Academy App: Apple | Android