For Imperfection: Why the Best AI Tools Will Be the Ones That Get Things Wrong
There's a race happening in every industry right now, and nobody's questioning whether the finish line is in the right place.
More pixels. More precision. More fidelity. More decimal places. The assumption baked into every tool, every update, every product roadmap is the same: better means closer to perfect.
I think that assumption is wrong. And I think AI is about to prove it.
I co-founded an architecture studio. Ran it for fifteen years. In that time I watched the tools go from hand-drawn sketches on tracing paper to photorealistic CGI renders that could pass for photographs. Every generation of software promised the same thing: more accuracy, more detail, more control.
And something got lost along the way.
The sketch on trace had life in it. It had the architect's hand, their hesitation, their confidence, their thinking process visible in every line. A client looking at that sketch wasn't just seeing a building — they were seeing a conversation. Possibility. Direction. Something they could respond to, push back on, get excited about.
A photorealistic render doesn't do that. It says: this is what it will look like. Full stop. There's no room in it. No conversation. No fingerprint. It's precise and it's dead.
I'm not saying renders are useless. They have their place. But the industry's obsession with chasing ever-higher fidelity missed something fundamental about how people actually respond to creative work.
They respond to the imperfections.
The rough sketch that captures the feeling before the details. The hand-drawn line that's slightly off but somehow more honest than the computer-generated one. The first take of a song that has more energy than the polished studio version. The Polaroid that makes you feel something the iPhone photo doesn't.
Imperfection is the signature of human involvement. It's how we recognise that a person was here, thinking, feeling, making decisions. Strip that out in pursuit of technical perfection and you strip out the thing that made it matter.
The Tool Paradox
Here's what every software company gets wrong: they assume their users want more precision. So every update adds more control, more options, more ways to fine-tune the output toward some theoretical ideal.
But ask any architect what they actually want from their tools and you'll hear the same thing: I want it to get out of my way so I can think.
They don't want a hundred sliders. They want to explore ideas quickly, see possibilities they hadn't considered, and make decisions based on feel as much as data. The best design moments don't come from precision — they come from accidents, surprises, unexpected juxtapositions that the designer's eye recognises as right even though nobody planned them.
The tools that chase perfection are actively working against this. They're optimising for output quality while killing the creative process that produces the best work.
AI Could Be Different
This is where it gets interesting. AI doesn't have to be another tool in the perfection race. It could be something fundamentally different.
A rendering engine gives you exactly what you specified. No more, no less. It follows the decimal. It doesn't have opinions. It doesn't surprise you. It's a precision instrument, and precision instruments are only as good as the instructions they're given.
AI doesn't work like that. AI interprets. It makes connections. It suggests things you didn't ask for. It gets things slightly wrong in ways that are sometimes more interesting than getting them exactly right.
Most people see that as a flaw. The hallucination problem. The lack of reliability. The reason you can't trust AI to be accurate.
I see it as the feature.
Not the hallucinations themselves — nobody wants factually wrong information. But the underlying quality that produces them: the ability to make unexpected connections, to see patterns across domains, to offer a perspective that wasn't in the brief.
That's what a good collaborator does. A good collaborator doesn't give you exactly what you asked for. They push back. They suggest alternatives. They see the problem differently. They bring something you didn't know you needed.
The best AI tools won't be the ones that execute your instructions perfectly. They'll be the ones that challenge your instructions productively. That treat the rough edges and the unexpected outputs as creative fuel rather than errors to eliminate.
The Paradox
Here's the thing that makes people's heads spin: I'm using AI to bring humanity back into the process.
That sounds contradictory. AI is supposed to be the thing that removes the human element. Automates. Optimises. Replaces the messy, inefficient, error-prone human with clean, reliable, scalable machine output.
But that framing assumes the human element is the problem. That the mess is what needs fixing.
What if the mess is the point?
What if the rough edges, the unexpected angles, the flaws and the surprises — what if those are exactly what makes creative work interesting, what makes tools useful, what makes the output worth caring about?
AI can be the tool that reintroduces those qualities at scale. Not by being inaccurate, but by being genuinely generative — producing options, perspectives, and combinations that a precision tool never would. Not replacing human judgment but giving it more to work with. More raw material. More possibilities. More of the creative friction that produces the best work.
We're not trying to build a perfect tool. We're trying to build a useful one. And useful, in creative work, means surprising, responsive, and comfortable with imperfection.
The flaws are where the interesting stuff lives. In people. In buildings. In software. In everything worth caring about.
The hundred-thousand-megapixel camera takes sharper photos. It doesn't take better ones.
Dave Mattu is a Welsh-British technologist and reluctant philosopher based in Huelva, Spain. He builds AI-powered tools for architects at skalm.space and writes about the space between humans and machines at davemattu.com.