Last week I was closing a national conference for a room of mortgage brokers, and I put a question to them about a job they all do.

You have to tell a client their plan isn’t going to work. Not a small correction. The real one, where somebody has made a decision about their life and you have to say it won’t hold. Is that a task for a human, for a machine, for AI, or for some combination of the three?

Hands went up for human, straight away, and I stopped there.

I didn’t go any further because they’d got it, and there was nothing I could add that wouldn’t have taken something away. It was the third task I’d put to them in about four minutes.

They knew where the line was. Nobody had drawn it for them.

Three tasks, four options

Here is the exercise exactly as it ran, because you can run it yourself without me.

I don’t look at a job as a job any more. A job is a stack of tasks. Answering the phone, chasing a document, checking a number, making a call about a person. Little bite-sized things that collect into one person’s name on an org chart, and once you look at the tasks instead of the title, most of the anxiety about the future of work stops being general and becomes answerable.

So take one task at a time and ask four questions of it. Is this better done by a human? By a machine, meaning the CRM, the record keeping, the deterministic stuff that lives inside the system? By AI? Or by a combination?

Task one. Collecting and chasing a client’s documents. Human? Very few hands. Machine, just let the system hold it and log it? Very few. AI? A few more. Combination? The room went up.

Task two. Deciding which lender to recommend, after the interviews and the due diligence are done. Human, all of it yourself? A few. A machine, pump the details in and hope? Very few. Hand the client’s whole file to AI and let it decide? And then combination, and up they went again.

Task three. The conversation above. Human, and I stopped.

Three tasks. Three different answers. No framework on the screen, no worked example, no preamble about maturity models. A room of people who spend their days on compliance and lending sorted their own work into human, machine, AI and combination, and got all three right in about four minutes.

That matters for a reason bigger than the exercise.

Every organisation I work with is having a large, abstract, expensive conversation about AI. Strategy sessions. Governance frameworks. Roadmaps with horizons on them. Underneath all of it the actual work is made of tasks, and the people doing those tasks already have a considered view about which ones are theirs.

Nobody has asked them. That’s the whole gap.

Why we defend the wrong thing

I was in fourth form, sitting at the back of my maths class, next to another boy also called Morris. Two of us, same name, same row. He became my best mate and it was enormous fun for everybody except the teacher.

I was a nerd and a geek years before Bill Gates made either of those a good thing to be. At the time it was a death sentence.

My father came back from Hong Kong with a little box in a brown faux leather case. It had a lid that came over the top, and inside were these magic buttons, and if you pressed the magic buttons it did the sums my maths teacher wanted from me. The sums I couldn’t do. The ones I could only pretend to do.

So I brought it in and put it on my desk, and for about ninety seconds I was going to be a geek who actually knew numbers.

Mrs Ruben was not having any of it. I can still hear her, which is why I can still tell you about it. She threw me out of the room and told me never to bring a calculator into her classroom again, because my mind was going to porridge and I needed to learn all of this properly.

I’ve been telling that story from stages for a very long time and the reason has never changed.

You and I don’t even use a calculator now. We went past them to spreadsheets, and past spreadsheets to AI, and at no point along the way did anyone hold a meeting about it. They became tools. Nobody argues about them.

Mrs Ruben wasn’t stupid and she wasn’t lazy. She was defending something real, which is the fear that a person will stop being able to do a thing and nobody will notice until it counts. I hear the same fear in boardrooms now, about AI rather than arithmetic, and it deserves a straight answer rather than a laugh.

She was defending the wrong item, that’s all. The valuable part of maths was never the arithmetic. It was knowing which sum to do and what the answer meant, and that part was never under threat from a box with a lid on it.

If you’re still fighting the fight the way she fought it, you’re going to lose. Not because AI wins. Because fighting is the wrong contest, and there’s a far better one on offer.

That’s not a position I’ve come to recently. I said in Perth in 2016 that it was never humans versus machines, and the trouble since has never been the machines arriving. It’s that almost nobody built the human half. This is the same argument, one task at a time.

It comes from watching people, not machines. We have looked for machines since somebody first put a stick under a rock. We have spent our entire history working out what to hand over and what to keep. We have never once got there by banning the object.

We get there by naming the work.

Human And

HUMAND is Human And. That’s the whole idea sitting inside the name, and it’s why I use the word rather than a diagram. Not human or machine. Not human versus AI. Human and.

The letters carry the method.

Horizon scan. Audit the task. What is it actually for, what are the constraints, what does a good result look like.
Unpack. Break the task into its human, machine and AI components, because almost nothing is purely one.
Mobilise humans. Decide what kind of human this needs. Full time, part time, a specialist for two days, someone remote, a blend.
Allocate machines. Put machines and automation where they genuinely add precision, consistency and endurance.
Navigate AI. Bring AI in for decision support, for the analysis, for the augmentation, with a person reading what comes back.
Design and deploy. Orchestrate the combination, then watch it and change it, because the right split this year is not the right split next year.

That last point is the one people miss. This isn’t a one-off audit that produces a laminated answer. It’s a running decision that moves as the technology, the skills and the business move.

And the reason I call it the word that changed humanity, which is a big claim and I’ll stand behind it, is what happens when machines and AI carry work together. It releases the brawn and the brain we used to have to put into getting something finished. That capacity goes back to the human, and the human goes back to what only a human can do.

Walk through any organisation and ask people about their work and you hear a version of the same thing. I love my job. I love the end result. I love the people. I just don’t love the tedious, repetitive parts of it.

The jobs disappearing are, overwhelmingly, the repetitive ones. That’s not nothing and I’m not going to pretend otherwise. But it is the most solvable problem in this whole subject, because the repetitive part was never the part anybody wanted.

Machines and AI are not the same thing

Machines and AI sit in separate letters on purpose, because most organisations run them together as one word and that single habit causes more governance trouble than any tool ever has.

A machine is deterministic. Same input, same answer, every time. When it’s wrong it’s wrong the same way for everyone, forever, until somebody changes the rule.

AI is probabilistic. Same input, possibly a different answer. Its confidence and its accuracy are different things. It fails unevenly, and it fails quietly.

Those need different checking and different lines drawn around them. Collapse them into “the tech” and you end up applying serious scrutiny to a chatbot somebody licensed in March while a scoring rule written in 2014 quietly decides who gets a payment plan.

The brokers made that distinction unprompted. When I offered “machine, just let the CRM hold it”, very few hands went up on any of the three tasks. They know the difference between a system that records and a system that judges. They live with both.

Somebody went home and did it

Kelly, a finance broker who was in that room, went away and put her own task through the tool, then posted what came back.

Her question was “annual reviews on client home loans”. The answer was 55% human accountability, 25% machine automation, 20% AI augmentation, with the reasoning attached: machines can pull the rate comparisons, calculate the refinance savings, check whether the loan structure still matches the client’s circumstances and flag whether they’ve drifted into a worse product tier. That quarter of the task is worth doing well because it’s accurate and consistent. The rest sits somewhere you can’t automate through without losing the relationship.

Her line about it was this. “I know I am still needed at 55%!”

That’s the argument of this piece, written by somebody who isn’t me, about her own work, the day after. She didn’t come away worried about being replaced. She came away with a number for the part that’s hers, and a clear view of what to hand over so she can spend more time on it.

Naming the work doesn’t shrink people. It tells them where they count.

The room that went somewhere harder

I ran the same exercise a couple of months back in Brisbane, with a Vistage group of chief executives, and it went straight to the place the brokers didn’t go.

I’d been talking about Lynn. In the 1980s, when I was starting out, Lynn was a CEO’s executive assistant who took shorthand at 120 words a minute. She’d sit in a board meeting with a pencil and a page, follow the whole debate, capture every word, and have the minutes typed before the board had left the room. She was extraordinary at it and she was paid accordingly.

I asked the room how many of them hire a stenographer now.

One of them answered before I could: “Yeah, but now? She’s absolutely useless.”

Blunt, and not wrong about the role. AI takes the note, synthesises it and sends it out before anyone reaches the lift.

I’m not going to dress that up, and I didn’t dress it up in the room either. We shouldn’t give people the false comfort that nobody loses a job, because the moment AI could take the note, Lynn’s job was gone. Some roles do go. Pretending otherwise insults the people standing in them.

Then somebody asked the question straight out. How do I use this to get rid of people?

It’s a fair question and it’s the one everybody is actually holding. I told them it was a perfect question, and asked them to park it for ten minutes, because you can’t think clearly about a task while you’re managing the anxiety attached to it. That anxiety has a name in my work, PTFA, past trauma and future anxiety, and it will happily eat an entire strategy session if you let it in early.

So we parked it, and they took a task apart cleanly. What came back was a roster problem. Three subcontractors don’t show up and the work still has to be done. Who’s available, who’s over hours, what does it cost, what can be dropped. They put the request handling on the machine, the analysis on AI, and the human at the end, reading the report and making the call about people.

Same shape as the brokers. Different sector, harder room, and they got there while the anxiety was on the shelf rather than in the chair.

The provocation on their worksheet was one I keep coming back to. Soon you may have to justify why a human is in each step. Not the other way round.

That’s the reason to do this now, while you still have the standing to decide what happens to the pieces.

The numbers say what the rooms said

Bring the data in now, and read it the right way round.

PagerDuty’s 2026 research, run by Wakefield across 1,250 office professionals, found 66% have used AI tools at work that nobody approved. Thirty-nine per cent would rather use AI and not mention it. Forty-eight per cent are uncomfortable telling their manager they used AI for an ordinary task.

Then the Australian half, which is the one that should hold your attention.

Treasury’s advice to the Treasurer, reported at the end of August, found that two thirds of Australian businesses say they’ve adopted AI in some form, and fewer than 10% would call that adoption significant. Wide and shallow. Treasury’s own concern is the gap between automating a task and redesigning the work, and it names the difference plainly: the benefit needs investment in data, systems, skills and organisational design, not just software licences.

Two thirds in. Fewer than one in ten of those actually changing anything.

In the same season APRA wrote to banks, insurers and superannuation trustees after a targeted review and said governance, risk management, assurance and operational resilience are not keeping pace with the scale, speed and complexity of AI adoption. It found AI in every entity it looked at, well past chatbots, into software engineering, claims triage, loan application processing and fraud disruption. Not experiments. Operational, customer-facing work.

You can read all of that as a compliance problem. Most of the commentary does.

Read it the way those rooms read it and two thirds of your people have already taught themselves a new capability, unprompted, unfunded, on their own time, with no change program and nobody asking them to. Every organisation I work with says it wants an adaptive workforce. You have one. It’s in the building, it has been experimenting for two years, and it has formed views.

The reason those views haven’t reached you isn’t secrecy. It’s that the question has never been put in a form anyone can answer. “What’s our AI strategy” is not answerable by a person in the middle of the work. “Is chasing this document better done by you, the CRM, AI, or some combination” is answerable in about four seconds, and everybody in the room has an opinion.

Ask the answerable question and the information arrives on its own.

The tasks that stay yours get sharper

In March 2018 I did a five-city tour for Choice, and in the write-up that followed I’m on the record calling mortgage brokers Finance Futurists, and telling them to think about their role as assessing and providing for somebody’s future with every loan they write. My own note from the time has the rest of it: “we’re in a people’s business and people love getting advice from a person they can see and trust. Keep humanity at the front and centre of everything you do.”

That was written before the Royal Commission and before anyone in that industry was talking about AI. I made the same argument in a room of brokers eight years on, and it lands harder now, because now there’s a machine that can do the processing.

When you sort the tasks honestly, the human column doesn’t shrink into a residue of whatever the machine didn’t want. It sharpens. The judgment becomes the product, because it’s no longer buried under the routine that used to eat the day.

The conversation I opened with proves it better than I can argue it.

In 1995 my wife was pregnant with our second child, our house was far too small, and I gave myself four months to build a bigger one. Builder ready. Plans ready. No money.

Norm was our bank manager at the ANZ, out on the country road, and Norm and Vera were our best friends. He knew everything about us before we were through the door. He sat me down and told me, to my face, that he loved me to death and it was not going to happen in the time I had. He couldn’t have emailed that. Email didn’t exist and we were still on fax machines. He had to say it to me in a room, and he did.

Then an ad came up for something called a mortgage broker, which in 1995 wasn’t in the Yellow Pages because it barely existed. I rang the man and told him the whole story. The house got built. My daughter arrived. The loan that couldn’t happen, happened, because somebody in a trade nobody had heard of decided it could.

Two conversations, thirty years apart from where I’m standing now, and neither of them was about paperwork.

A room of brokers, given four options and no guidance, put that kind of conversation in the human column without a moment’s hesitation. They know exactly what they’re for. They always have.

That’s the good news in this whole subject and almost nobody is saying it. People are not confused about what they’re for. They’re waiting to be asked one task at a time.

Three rooms, nine days, one question

It’s been a full stretch. Three major sessions inside nine days, two conference keynotes and a three-hour leadership workshop, across three sectors that share almost nothing, alongside everything else running beside them.

The brokers, where the human column was the conversation about somebody’s future.

The day before, at The Star on the Gold Coast, Queensland’s civil contractors at their state conference, on the decade that will test civil construction. Sensors, drones, machine control, digital twins, and what arrives after 2032. Leanne, at the federation, wrote afterwards and put the human column better than I did from the stage: “the importance of retaining experienced workers and transferring their knowledge as workforce roles and required skills evolve.” She also named why it landed. “It did not treat these trends as distant possibilities. Many are already being experienced across Queensland’s civil construction industry.”

There’s a hard practical edge in her sentence. If an experienced operator’s judgment is now partly carried by a machine and nobody has said which part, then transferring their knowledge is a phrase with nothing attached to it. Sort the tasks and you can name it. Name it and you can teach it. A succession problem solved by a conversation rather than a system.

And before both of those, three hours with a leadership cohort at Holmesglen, an institution I’ve worked with for more than thirty years across teaching, consulting and a long run of workshops. That length of relationship is its own kind of instrument. You get to watch one organisation’s thinking move over decades rather than over a lunch.

They ran the same task sort, and one pair came back with a tiered model for student services. Humans as the ultimate decision-makers. The student management system as the machine, holding and distributing the data. AI doing the analysis. And then a human reading the AI output before anything moved, because they’re mindful of the hallucinations AI can produce. Four layers, in order, with the check built in. I told them it was a beautiful melody, because that’s the harmony of the three working properly.

In education the live nerve underneath all of it is the credential. If the qualification was built around the part a machine now does comfortably, what’s the qualification for? It isn’t finished. It moves, to the judgment and the practice that were always the valuable half and were never the part that got examined.

Different sectors, different rooms, same question.

What’s still ours to do, and who says so?

Try it on one task, today

Not a program. Not a steering group. One task.

Pick a role you know well and write out its tasks. Not the position description, the actual tasks, the bite-sized things that fill a day. Then take them one at a time and ask the four questions of each. Person. System. AI. Some combination of them.

Most tasks come back as a mix, and the mix is the useful part. A task that’s 70% system and 30% person tells you far more than a task filed under “automate”.

You’ll get through most of a role inside an hour, and the arguments you have over the awkward ones are the most useful hour your leadership team will spend this quarter, because those are exactly the tasks your people are already deciding about without you.

Then say what happens to the time. Out loud, in writing, before anybody asks. If you haven’t worked that out yet, say that instead. People can work with an honest unknown. They can’t work with silence, because silence gets filled in with the least generous version available.

Then ask the room, one task at a time. There’s no right or wrong answer to any of it. It’s a question about work style, about what suits the work and what suits the business, and nobody should feel intimidated by it.

And take this with you, with my compliments for reading this far. I built humand.life because I talk about this constantly and I wanted to see whether AI could make the call itself. Put a task in and it breaks it down and gives you the split, the way it did for Kelly, with the reasoning attached. Probably a good combination rather than the best one, because this is individual to your business and it should be. It’s free. There are no steak knives.

Not prediction. Preparation.

Mrs Ruben and I were arguing about the wrong thing, and neither of us knew it at the time.

She thought the question was whether the calculator belonged in her classroom. It was never that. The question was which part of the maths problem was mine to work out, and which part was the machine’s to do for me. Nobody had ever separated those two, so we argued about the box instead.

That’s the same question sitting in front of you now. It hasn’t got any harder in fifty years. It has only got more expensive to leave unanswered.

Take one task. Break it into what it’s actually made of. Work out what a person should do, what the system should do, what AI should do, and where two or three of them work together. It won’t come back as a tidy split. Kelly’s came back 55, 25 and 20, and that’s what a real answer usually looks like.

Then say what happens to the time.

The organisations that come through this well won’t be the ones with the most detailed AI usage policy or the biggest enterprise licence. They’ll be the ones where somebody stood up in front of their own people, asked the question in a form a person could actually answer, and then listened to what came back.

Your people already know where the line is. Ask them.

Choose Forward.

---

Morris Misel is a foresight strategist and business futurist. He delivers opening and closing keynotes, and runs leadership workshops, for boards, executive teams, associations and conferences across every sector. HUMAND is Human And, his framework for deciding what work is best done by humans, machines, AI, or a combination. Try it free at humand.life.

Sources: PagerDuty 2026 Shadow AI research, Wakefield Research, 1,250 office professionals · APRA letter to industry on artificial intelligence, 30 April 2026 · Treasury advice on AI adoption, ABC News, 31 August 2026 · Australian Broker, Rebecca Pike, 18 April 2018 · Kelly, LinkedIn, September 2026

Frequently asked questions

What decisions should AI make versus humans?

There is no single answer at the level of a job, which is why the question feels unanswerable. It becomes answerable at the level of a task. Take one task at a time and ask whether it is better done by a person, by a machine such as a CRM or record-keeping system, by AI, or by a combination. Most tasks come back as a mix rather than a clean category. Morris Misel’s HUMAND framework is built around exactly this task-level sort.

What is the HUMAND framework?

HUMAND is Human And. It is Morris Misel’s framework for deciding what work is best done by humans, machines, AI, or a combination, and the letters carry the method: Horizon scan the task, Unpack it into components, Mobilise humans, Allocate machines, Navigate AI, then Design and deploy the combination and keep changing it as conditions move. Machines and AI sit in separate letters deliberately, because a machine is deterministic and fails the same way for everyone, while AI is probabilistic and fails unevenly and quietly.

Why do employees use AI tools their employer has not approved?

PagerDuty’s 2026 research found 66% of office professionals have used AI tools at work that were not approved, and 48% are uncomfortable telling a manager they used AI for an ordinary task. Treasury’s 2026 advice found two thirds of Australian businesses have adopted AI in some form but fewer than 10% would call that adoption significant. The gap is not secrecy. It is that nobody has put the question in a form a person doing a job can answer. Ask about a specific task and people answer immediately.

How do you run an AI task audit on a role?

Pick one role and write out its actual tasks rather than its position description. Take them one at a time and ask four questions of each: person, system, AI, or some combination. Most of a role can be sorted inside an hour, and most tasks come back as a mix. Then state in writing what happens to the time that is freed up, because that answer determines whether people engage honestly. The free tool at humand.life performs the same sort on a single task and returns a percentage split with reasoning.

Is AI reducing the value of professional judgment?

The evidence from rooms points the other way. When tasks are sorted honestly the human share stops being the residue left after automation and becomes the identifiable product. A finance broker who ran her own annual client loan reviews through humand.life received a 55% human accountability, 25% machine automation, 20% AI augmentation split, and her response was that she is still needed at 55%. Naming the split raises the visibility of judgment rather than lowering it.

Does AI mean some roles disappear entirely?

Yes, and pretending otherwise insults the people standing in them. Morris Misel uses the example of the stenographer: once AI could take and synthesise meeting notes, that role was gone. His position is that organisations should not offer false comfort, and should instead sort tasks while they still have the standing to decide what happens to the pieces. The provocation he puts to leadership groups is that soon organisations may have to justify why a human is in each step, rather than the other way round.