A companion to Design is not UI
It's OK to be human.
AI is not your replacement.
AI is the most powerful tool we've ever handed a person. It still can't tell you what's worth doing, or whether it worked.
It has changed how fast we can work. It writes rough drafts, digs through research, and clears the repetitive slog that used to eat whole weeks. No generation before ours had a tool like it.
And yet the loudest question in the room is the wrong one.
Scroll to see where the human fits.
•5 min read
The AI question
Everyone is asking the wrong question.
The question you hear everywhere right now is: will AI replace me? It's the wrong question. It treats the work as a single seat, and the only fight left is who gets to sit in it. No tool has ever worked that way.
The better question is quieter. What does the work look like when a person and a machine do it together? Which parts are yours, which parts are the machine's, and who answers for the result when the two of you disagree?
Replacement is a story we tell because we're scared. The truer story is duller: you and the machine end up side by side, each taking the part you're better at. Guess which one gets shouted from every stage.
The same ditch, dug two ways. The shovel doesn't dig. It turns a week of digging into an afternoon.
The shovel
Think of it as a shovel.
A shovel is a wonderful thing. Hand one to someone with a ditch to dig and the work that would have taken a week takes an afternoon. But set the shovel down in the field and come back tomorrow: the ditch is exactly as deep as you left it. A shovel digs nothing on its own. It has no idea where the ditch should go, how deep, or why you're digging at all.
Now take the shovel away and leave the person. They'll still dig the ditch. It just costs them ten times the sweat and a week they didn't have. They know exactly where it goes and why. What they're missing is the thing that gets them there fast.
Neither one finishes the job alone. The shovel without a hand is a piece of metal in the dirt. The hand without the shovel is a slow, aching week. Put them together and the ditch is dug by lunch.
AI is the shovel. You are the hand that knows where the ditch goes. Don't mistake the tool for the digger, and don't dig with your bare hands when there's a shovel leaning against the wall.
Every dot is a piece of the job. Most of them the tool can take. The red one needs a person.
Not a replacement
A co-laborer, not a stand-in.
Treat it as a replacement
Hand it the whole task
and hope the output is right
Speed is the only goal
faster is always better
Nobody owns the result
the tool did it
Judgment is a bottleneck
something to automate away
Treat it as a co-laborer
Hand it the heavy lifting
and keep the direction yourself
Speed serves a purpose
faster toward the right thing
You own the result
the tool has no skin in it
Judgment is the point
the part only you can do
A co-laborer does the work you point it at. What it won't do is decide whether the work is worth doing, or answer for how it turns out. That was your job before AI showed up, and it still is.
Hand it the whole dig and the ditch drifts. Keep your hands on the work and it builds on itself, all of it pointed where you aimed.
Two runs of the same mistake. On its own, it feeds everything built after it. With a person in the loop, it gets caught while the fix is still cheap.
Unregulated, it compounds
Left unchecked, mistakes multiply.
AI makes mistakes. So do people. That's not the interesting part. What's interesting is what happens to a mistake nobody catches.
A person working alone catches most of their own errors, because they're slow enough to notice. AI is not slow. It can take one wrong assumption and build a hundred confident things on top of it before anyone looks up. The error doesn't sit still. It seeds the next output, and the one after that, until you're not fixing a mistake anymore. You're unwinding a whole worldview.
This is true of people too, given enough speed and no one checking. Put a person on a task with a hard deadline, no review, and no one to answer to, and the mistakes compound the same way. Speed without a regulator is how both of you go wrong. The machine just gets there faster.
The regulator is the human in the loop. That means being inside the work, close to the moments where a wrong turn is still cheap to fix. Watching from a distance doesn't count. You have to actually do the parts that are yours.
AI amplifies Groundwork and Production. Direction, Judgment, and Accountability stay with you.
Apply it correctly
Point it at the right work.
Used well, AI doesn't replace any step in that loop. It speeds up two of them. Just two. The rest stay slow and human on purpose.
01
Direction
Human only
Deciding what to build and why it matters. Nothing to hand off here.
02
Groundwork
AI assists
Research, summaries, first passes over material you would otherwise read line by line.
03
Production
AI shines
Drafts, variations, boilerplate: the long repetitive middle. The tool never gets tired, and a draft is cheap to redo.
04
Judgment
Human only
Is this actually good? Does it serve the goal? The tool cannot feel the answer.
05
Accountability
Human only
When it ships, your name is on it. Never the tool's.
Give AI the long, grind-it-out parts. The digging. Keep the parts that need a human to know where the ditch goes, and whether it will hold water.
But that line isn't as tidy as it sounds. The tool is uneven in ways you won't see coming. It will nail one hard job and fumble the next one that looked just like it, and sound every bit as sure both times. There's no clean list of what it's good at. The only way to find its edges is to stay close enough to feel them.
Work this way and you stop racing the machine at all. You move at a pace neither of you could hit alone. It only works while a real person keeps doing the human parts on purpose. The day they stop, it quietly falls apart.
Two ways to get it wrong
Don't fear the shovel. Don't misuse it either.
There are two ways to get this wrong, and they look like opposites. One is fear: refusing the tool, digging by hand out of principle while the ditch you could have finished by noon eats the whole week. The other is misuse: handing the tool the entire job, walking away, and being surprised when the ditch runs the wrong direction.
Both come from the same misread of what the tool is for. Fear sees a threat. Misuse sees a substitute. It's neither. A shovel never dug a ditch on its own. Someone has to pick it up and aim it.
So learn what it's good at, and learn where it goes blind. Learn your own part cold too: the judgment, the taste, the ownership no tool has ever carried for anybody. Then pick it up and dig.
Neither of you is enough on your own. You were never supposed to be.
What this is (and isn't)
This is not anti-AI.
We build with these tools every day. They make the work faster and take the tedium out of it. Swearing them off doesn't make you noble. It just makes you slow.
This site is written by people who reach for the shovel, and reach for it often.
This is pro-human.
It's an argument for keeping a person in the loop who is actually working it: setting direction, exercising judgment, answering for the result. Presence isn't enough. The human parts still have to get done. Sitting near the work isn't doing it.
This site is:
- +Pro leverage
- +Pro speed with a direction
- +Pro humans doing the human parts
- +Pro owning the outcome
This site is not:
- −Anti-AI
- −Anti-automation
- −Anti-speed
AI extends what a person can do. It was never going to be the person.
In closing
It's OK to be human.
The machine is faster than you, and it always will be. Fine. A shovel moves more dirt than a pair of hands too, and nobody ever stood next to a shovel feeling replaced.
What it can't do is care whether the ditch was worth digging. It can't pick the spot, and it can't put its name to the result. That work stays on your desk. It isn't a leftover. Deciding, and standing behind what you decide, is the actual job.
Pick up the tool. Stay in the loop.
AI is leverage, and we reach for it every day. But judgment isn't something you can hand off to it. Judgment is the hand that decides where all that force gets aimed.
If this landed, send it to someone who needs to hear it.
The companion piece
Design is not UISame point of view, aimed at the work instead of the person. If you nodded along here, that one is the other half of it.
Further reading
- Erik Brynjolfsson: The Turing Trap on why building AI to replace people is the trap, and building it to augment them is the way out.
- Ethan Mollick: Centaurs and Cyborgs on the Jagged Frontier on splitting the work with the machine, and learning the hidden line where it quietly stops being any good.
- David Autor: Applying AI to Rebuild Middle-Class Jobs on AI that widens who gets to do expert work, instead of hollowing out the people who do it.
- Simon Willison: AI-Enhanced Development Makes Me More Ambitious on a working programmer taking on projects he once shelved, while keeping the direction and the debugging for himself.