Faster Garbage Is Still Garbage
The AI debate keeps circling efficiency and authenticity. The question that matters is a different one.
Open any AI conversation right now, in agencies, in startups, in corporate town halls, and you will hear the same two words within minutes: efficiency and productivity.
We save 40% of production time. We ship three times as many assets. One person does the work of five.
Then, usually in the second half of the panel, comes the counter-movement: authenticity. Is it still real? Is it still human? Can we trust it? I wrote about that myself, and I think the question is legitimate.
But both debates share a blind spot. They discuss how things are made and who made them. They almost never ask the more uncomfortable question:
Is what I’m making actually new, or am I producing the same old rubbish, only faster?
Speed is a multiplier, not a direction
Efficiency is a multiplier. It scales whatever you feed into it.
Feed it a sharp idea, and you get that idea into the world sooner and in more variations. Feed it a stale idea, and you get stale at industrial scale. A mediocre campaign concept does not improve because the key visuals took two hours instead of two weeks. A forgettable article does not become relevant because it was drafted in 40 seconds.
Productivity metrics cannot see this. They measure output per unit of input. They are blind to whether the output was worth producing at all. A team can double its throughput and halve its impact, and the dashboard will show a triumph.
This is the oldest trap in industrial history: optimizing the process of something that should have been questioned first.
Why AI makes the trap worse
Generative models are, by design, machines of the plausible. They are trained on what already exists and are very good at producing what looks like it belongs. That’s their strength. It is also the risk.
Left on default settings, with a generic prompt and a deadline, a model returns the statistical middle: the typical headline, the typical moodboard, the typical “authentic brand story.” It feels like progress because it is fast and polished. But polish is not novelty. The result is often competent, coherent, and interchangeable.
When everyone has access to the same machines, and uses them the same way, the market does not get richer. It gets smoother. Average becomes the cheapest thing in the world, and therefore worth almost nothing.
The authenticity debate is a detour
Authenticity matters. But it can become a convenient place to hide.
I’ve been here before. In The Perennial Question of Imaging Techniques in Marketing — Which AI Is Making Worse, I looked at the old choice between photography, illustration, and AI visuals. My conclusion: the technique was never the real question. Clarity was. “Authentic” stock photography became its own genre of obvious artifice long before AI arrived, and generic AI aesthetics are the same strategic failure with a different production method. Speed without clarity, I wrote there, is creative procrastination with better-looking output.
This article takes that thought one step further, from images to everything we produce with these tools.
If the question is only “was this made by a human?”, then a human who recycles the same clichés passes the test, and a thoughtful person using AI to explore something unexplored fails it. That makes no sense. A tired idea made by hand is still a tired idea.
Get Björn Burkhard’s stories in your inbox
Join Medium for free to get updates from this writer.
The better test doesn’t ask about the origin of the work. It asks about its effect:
- Does it show me something I haven’t seen?
- Does it change how I think about the problem?
- Would anything be missing if it didn’t exist?
Authenticity without novelty is nostalgia. Efficiency without novelty is noise.
Four questions before you press “generate”
I use a short check before committing resources, whether that’s compute, budget, or my own attention. None of it is complicated. It is just easy to skip when the tool is fast.
1. What is the new thing here?
Name it in one sentence. If you can’t, you’re probably about to reproduce a reference. Not a verdict, but a warning.
2. What would the default output look like, and am I different from it?
Run the obvious prompt first, on purpose. Treat the result as the baseline you have to beat, not as the deliverable.
3. What do I bring that the model doesn’t have?
Taste, context, a point of view, access to a real person or a real problem, a decision someone else wouldn’t dare make. If the answer is “nothing”, the tool is doing the thinking, and so is everyone else’s tool.
4. Is this worth the cost?
Time, money, energy, and the reader’s attention are all resources. Cheap to generate does not mean cheap to publish. Every mediocre asset takes up space that a good one could have had.
Use the speed where it helps thinking
The point is not to slow down. Speed is a gift when it’s aimed correctly.
Use it to explore more directions before you commit. To kill weak ideas in minutes instead of weeks. To test the absurd, the risky, and the half-formed. To prototype what used to require a budget meeting. To get to the second and third idea, which is where originality usually lives, because the first one is almost always the one everybody has.
In that mode, AI doesn’t replace the creative decision. It gives you more material to decide with.
The failure mode is the opposite: using speed to skip the thinking. Producing first and asking later. Filling channels because they exist.
The real metric
If I had to replace the productivity dashboard with one number, it would not be output per hour. It would be something like: how much of what we made would we miss if it were gone?
That number is hard to measure and easy to fake. But it points in the right direction. It rewards ideas over volume, decisions over defaults, and work that earns its place.
The AI debate will keep asking whether it’s faster, and whether it’s real. Fine. Those are fair questions.
But the one that decides whether any of this was worth it is simpler, and older than any model:
Is it new, and does it matter?
If not, the machine didn’t fail you. It just helped you do the wrong thing, very efficiently.
