What to keep human
A working list for deciding which parts of your business should never be automated — and why the usual answer is backwards.
Most people draw the line in the wrong place.
They automate the writing and keep the admin. They let a tool draft the thing that carries their voice, then spend Thursday afternoon moving invoices between two systems by hand. It feels careful. It is exactly upside down.
Here is a better rule, and it has nothing to do with what the tools can technically do.
Keep the work where being wrong is expensive and being average is fatal. Automate the work where being wrong is cheap and being consistent is the whole point.
That single line will sort most of your business in about ten minutes. What follows is the longer version, and the part where it gets uncomfortable.
The four things that stay yours
Your point of view. What you believe about your field that not everyone believes. This is not a tagline and it is not "our values." It is the set of opinions that make some clients lean in and others quietly decide you are not for them — and both of those reactions are the point. A tool trained on everything will give you the average of everything, and the average has no opinions. It cannot have one; it is a summary of a million of them.
Taste. Not the ability to make something. The ability to look at nine versions and know which one is right, and to know it faster than you can explain it. This is the scarcest thing in any creative business and the least discussed, because it sounds arrogant to claim and impossible to teach. It is also, right now, the entire job. A machine will happily generate the nine. Choosing is what you are paid for.
Judgment under real stakes. Whether to take the client. Whether to say the difficult thing on the call. Whether this launch is ready. These are decisions with consequences that land on a person, and they should be made by a person who will carry them.
The relationships. Obvious, routinely violated. If a client can tell that the message they received was assembled rather than written, you have not saved twenty minutes; you have spent some of the only thing that makes you hard to replace.
The things that should have been automated a year ago
Now the other side of the line, which is longer than most people expect and considerably less glamorous.
The second draft of anything. Reformatting one piece of work into four shapes for four places. Turning a transcript into notes. Pulling the same five facts out of a document for the tenth time. Answering the question you have answered forty times, in your own words, correctly, again. Keeping two systems in sync. Producing the variant. Checking your own work against your own standards.
None of that requires taste. All of it eats the week that taste needs.
And that is the real argument for automating it — not efficiency. Time. The thinking that produces a point of view does not happen in the ninety minutes between two calls. It happens in the long, unhurried stretch you never get, because the repetition ate it.
The part everyone gets wrong
Here is the failure I see most, and it is not technical.
People automate the output without ever writing down the input. They point a tool at their business, ask it to produce, and are disappointed when what comes back sounds like everyone. Then they conclude the tools do not work for creative businesses.
The tools worked perfectly. They were given nothing to work from, so they used the average, which is the only thing they have by default.
A system with nobody's taste in it produces the average — and the average is precisely the thing you are trying to escape.
So the order is not negotiable. Write down what you believe, how you sound, who you are for, and what "good" means to you, in enough detail that a stranger could apply it. Then automate against it. The document is boring to make and it is the entire difference between a system that sounds like you and one that sounds like a competitor.
How to know you have drawn the line correctly
Three tests, in increasing order of honesty.
The stranger test. Show someone the output with your name removed. If they cannot tell it is yours, the foundation is too thin — the system is running on the average with your logo on top.
The wince test. Read what the system produced as though a competitor wrote it. If you would not have made that choice, do not ship it and do not shrug. That is information about your foundation, not a reason to lower the bar. Feed it back.
The muscle test. The uncomfortable one. After six months, are you thinking more clearly about your work, or less? If a tool has taken over the deciding as well as the doing, your own judgment gets quieter, the way a muscle does when a machine takes the weight. That is the actual risk in all of this — not that a machine replaces you, but that you gradually stop using the part of yourself that was never replaceable.
The shape of it
The businesses that come out of this decade well will not be the ones that automated the most. They will be the ones that were clearest about what they refused to automate, and ruthless about everything else.
Write the list. Two columns, half a page, an afternoon. Keep the things that only you can be wrong about. Hand over the rest.
Then go and use the time for the thing the time was for.
The Long Lunch
Human-led brands. Intentional AI systems. Book a free 15-minute call.