Stop Thinking in Roles. Start Thinking in Workflows.

Before we start. I'm opening 5 spots for my flagship done-for-you AI consulting program. This is where I build your AI consulting firm from scratch. The whole thing. Positioning, brand, website, content, messaging, outreach, and help actually fulfilling the first client work. Five, then it shuts. Book a call here

A job title is a container somebody drew around a stack of workflows. Open it.

September 19, 2026

Read Time: 9 minutes

Work is being rebuilt underneath us, and businesses that do not adapt are going to watch their margins compress while somebody down the road quietly doubles theirs.

I am not the only person saying this. It has become one of the more common conversations in the space this year and for once the consensus is broadly right.

But almost everyone is still framing it as a question about jobs. Which roles survive. Which are exposed. Which are safe.

Gartner put out its 2027 CIO survey this month and the advice was to redesign work before expanding teams, classifying every role as strategic, core, foundational or misaligned. Thorough piece of work. Still the wrong unit. (Gartner, CIO Planning for 2027)

Because a role is not a thing. A role is a container. It is a bundle of workflows somebody decided, at some point, one human should be responsible for, and the boundaries of that bundle were drawn by hiring budgets and history rather than by any logic about how the work fits together.

So when you ask whether a role is at risk from AI, you are asking a question that cannot be answered. The container holds ten or fifteen different kinds of work and AI is brilliant at some of them and useless at others.

The better question is much smaller. Which workflows in this business are worth automating, and in what order.

A Role Is Just a Container

Take a video editor. Sounds like one job. Everybody knows what it is.

Now open it.

Competitor research. Pure pattern recognition at volume.

• Set up a separate account so the algorithm feeds clean data rather than their own viewing history

• Work through a stack of competitor videos

• Save them into a folder

• Break each one down. What is the thumbnail doing, what is the title structure, what happens in the first three seconds

• Repeat across tens or hundreds of videos

Ideation. Taking a proven concept and generating five ideas off it that actually sound like your brand rather than somebody else's. AI assists this one heavily. The call at the end is still judgement.

Editing. Cutting sections, adding overlays, deciding what stays and what goes. This is craft, and a person is still meaningfully better at it than any system you could build.

Thumbnail production. Same discipline as competitor research, pointed at images, and the best version of it looks outside your own industry entirely.

There is a lovely example of this. A niche channel that does nothing but motorbike hill climbing pulls hundreds of millions of views, largely on the strength of how it frames its thumbnails. Several very large entertainment channels, nothing to do with motorbikes, took that framing and built videos around it.

That is a workflow, not a flash of inspiration. A consistent process for finding what is working visually, including well outside your category, and working out why.

Title testing. Structures, lengths, what earns the click in your niche specifically.

Performance analysis. What worked, what did not, and what the pattern is across the last thirty videos.

Split testing. Constantly running thumbnails and titles against each other, which is measurement work where frankly AI is better than most people.

Seven distinct workflows sitting inside one job title, and they have almost nothing in common with each other.

Three of them are pattern recognition at volume. Two are measurement. One is craft. One is judgement with heavy assistance.

Which is why "should we automate the video editor" has no answer. And "which of these seven, in what order" has an obvious one.

Nobody got into video editing because they love cataloguing thumbnails. You are not taking anything away from them.

What Makes a Workflow Worth Automating

Three things decide it.

Frequency. How often does this run? Something that happens forty times a week compounds. Something that happens twice a quarter almost never justifies the build, however tedious it is.

Human error rate. How easily does this get messed up when somebody is tired, rushed, or covering for a colleague on holiday? Repetitive work with high error potential is prime territory, because machines do not get bored and people do.

Severity. What happens if it goes wrong. And this one runs the opposite way to how people expect.

High severity is usually a reason to be careful, not a reason to go first. Finance and payments is the obvious case. You can automate the preparation, the reconciliation, the flagging of anomalies, the drafting of the report. You do not hand over authorisation of money movement. There is always a human verifying before anything irreversible happens.

So what you are hunting for is high frequency, high error rate, low severity. There are far more of those in any business than the owner expects, and most of them are boring, which is exactly why nobody has looked at them.

Where This Bites for an Owner

Here is the practical consequence, and I want to be precise because it is easy to misread.

Right now an owner has a gap. Output is not where he wants it. So he reaches for the only lever he has ever had, which is to hire somebody.

He writes a spec. We need a creative strategist. Then three months of recruitment, a salary, onboarding, training, and a hope it works out.

The workflow lens interrupts that with one question. What is a creative strategist, actually?

Break it apart. List every workflow that person would own. Then sort them into what AI can handle, what needs human judgement, and what needs genuine craft.

Very often eight of the twelve can be automated or heavily assisted. Which means he does not need a full-time creative strategist at all. He might need somebody fractional. Or he might realise that what he actually needs is a videographer, because the strategy layer can be systematised and the shooting cannot.

Completely different decision, reached in an afternoon rather than after three months of recruiting and a year of salary.

And there is a second-order effect here that matters more than the saving, and it is the thing that makes owners lean forward.

Finding good people is not expensive. It is hard. Good candidates are scarce, slow to onboard, and there is a real chance they do not work out. For a lot of growing businesses recruitment is not a cost problem, it is a ceiling.

Automating the workflow layer inside existing roles breaks that ceiling. The team you already have, who understand the business and know what good looks like, get their week back. In my experience that is worth something in the region of thirty percent more top line without adding a single person, and because there is no salary attached to it, a disproportionate amount of that lands as profit.

Same team, more output, higher margin, no recruitment cycle.

How to Say It Without the Room Closing

Be careful, because "think in workflows, not roles" is one short step from "we do not need these people", and the second a room hears that you have lost it.

I wrote three weeks ago about handling the replacement objection and everything in that issue applies here, harder.

The framing that works happens to be the honest one. You are almost always sitting with somebody who wants to hire. They have a gap and they are about to spend money filling it. So the conversation is not about removing anyone, it is about whether the next hire is actually the answer.

And for the people already there, the message is capacity rather than replacement. Removing the repetitive third of somebody's week is a gift to the person doing it.

There may well be a reduction in hiring. That is true and you should not pretend otherwise. But a reduction in hiring and a reduction in headcount are completely different things, and being straight about that distinction is what keeps the room open.

It Changes How You Scope the Audit

If you are running AI assessments, this is not a new methodology to bolt on. It is the methodology.

We do not map departments. A department is just another container, another line somebody drew. We map workflows, every one, across every role in scope.

That granularity is exactly what justifies a serious fee. Anyone can produce a document saying the marketing function could benefit from AI. That is worth nothing because nobody can do anything with it on Monday.

What is worth five figures is a map that says: here are the forty seven distinct workflows running in this business, here are the eleven that are high frequency and high error, here is what each currently costs in hours and salary, here is what it would cost to change, and here is the order to do them in.

That is a build plan rather than an opinion. It is the difference between a report that gets read once and a report that gets worked through.

Sometimes a whole workflow comes out end to end. More often you take the data gathering, the drafting and the analysis, and leave the judgement where it belongs. That nuance only becomes visible once you have broken the work down properly, which is the part nobody inside the business has ever done.

What Doesn't Decompose

Now the counterweight, because the whole argument falls over without it.

Judgement. You never outsource judgement. Not the final call, not the risk assessment, not the decision about whether a thing is actually right. AI prepares, analyses, models and recommends. A person decides. That line does not move.

Relationships. Trust is built between people. Somebody about to make a significant decision wants a human they know on the other end of the phone. You can automate the preparation for that conversation. You cannot automate the conversation.

Presence. On-site work, physical work, anything that needs a person to actually be somewhere. It is why the in-person model keeps winning and why Palantir built their entire approach around forward deployed people.

Reputation. Your name and your standing in a market. Earned through behaviour over time, and there has never been a shortcut.

One smaller thing worth saying. There are still tasks today where a person is simply faster than anything you could build, despite the task looking trivially automatable. Not clever tasks. Stupid ones. If you have ever spent a week building something to save four minutes, you know exactly what I mean.

That will probably not be true in a year. It is true now, and pretending otherwise gets you found out by anyone who has actually tried.

Is There a Limit?

I have thought about this a fair bit and I do not think there is a meaningful ceiling, as long as the judgement layer stays human.

While decisions, risk and accountability sit with people, I do not think a business can over-automate its workflow layer, because the workflow layer is by definition the mechanical part.

Where this goes is genuinely significant. Businesses are heading somewhere far more efficient than anything we have operated with before, and my own view is that margins across most industries improve dramatically as it plays out. Possibly something like fifty percent. That is a view rather than a forecast, but it is the direction I would bet on.

And the ones who get there first compound, because better margins fund faster reinvestment.

Every audit we have ever run has been this, incidentally. There is no standout case study because it is the entire job. Map the workflows, find the ones quietly costing money, build in priority order.

The Real Point

Everyone else is sorting people into categories. That is the old frame and it leads straight to a defensive conversation about whose job survives.

Go underneath it. A role is ten or fifteen workflows somebody once bundled into a person because they had to.

Break the bundle. Find the ones that run often, break often, and would not hurt anybody if they failed. Build those first.

Then stop asking whether AI is coming for jobs, and start asking the only question with an answer. Which work, and in what order.

See you next week,

– Andrew

P.S. Ready to build your own Vibe Consulting business? Book a 1:1 call here to see if you're a good fit for my personal done-for-you program.