This is part 3 of our series on The Democratisation of Building.
For most of its history, software has been an engineering discipline. You planned it on paper, built it brick by brick, tested it methodically, and hoped you’d specified correctly at the start – because changing direction halfway through was ruinously expensive. Software was steel and concrete: strong once set, but utterly unforgiving if you got the blueprint wrong.
That metaphor is breaking down. Rapidly.
Software is becoming clay. Something you shape with your hands. Push it, spin it, look at it, push again. If it’s not right, you don’t demolish and rebuild — you reshape. The material itself has changed.
The Loop
There’s a pattern emerging in how AI-assisted software gets built, and it’s almost comically simple. Some people are calling it the Ralph Wiggum loop (after the Simpsons character who just… does things, cheerfully, without overthinking).
It goes like this: spec → implement → test → commit → repeat. Each cycle starts fresh. You describe what you want. The AI builds it. You look at what you got. You adjust your description and go again.
It sounds almost too simple to be real. But the results are starting to speak for themselves.
Simon Willison — a well-known developer and one of the sharpest observers of AI tooling — recently described building a fully working HTML5 parser that passes over 9,200 conformance tests. He did this while decorating a Christmas tree. The AI worked unattended for four and a half hours, methodically implementing, testing, fixing, and committing. By the time the fairy lights were up, the parser was done.
That’s not a toy demo. An HTML5 parser is genuinely complex software, the kind of thing that would normally take a skilled developer weeks of focused effort.
Or take Cursor, the AI-powered code editor, which was used to build a functional web browser — over a million lines of code across a thousand files — in a single week. A web browser. The kind of software that used to be the exclusive domain of teams of hundreds at companies like Google and Mozilla.
These aren’t anomalies. They’re signals.
From Engineering to Craft
Something fundamental is shifting here, and it’s not just about speed.
When software was engineering, the core skill was implementation. Could you write the code? Did you understand the architecture? Could you wrangle the database, manage the servers, debug the memory leaks? The barrier to entry was high, and for good reason — getting it wrong had consequences.
But when AI handles implementation, the craft moves upstream. The core skill becomes specification. Can you describe what good looks like? Can you look at what was produced and tell whether it’s right? Can you articulate the difference between “technically works” and “actually solves the problem”?
This is the potter’s wheel metaphor we keep coming back to. A potter doesn’t engineer a vase from a technical drawing. They push the clay, watch it respond, adjust their pressure, and shape it through iteration. The skill isn’t in the material — it’s in the eye, the intent, the feel for when something is right.
That’s what building software is starting to feel like.
The Market Agrees
If you want evidence that this shift is real, follow the money. Lovable, an AI-first app builder, went from zero to $200 million in annual recurring revenue in twelve months. That’s not a rounding error. That’s an entire market screaming: we have things we want to build, and we’ve been waiting for a way to build them.
The demand was always there. What changed is the translation layer. The expensive, time-consuming, expertise-dependent process of turning “I wish we had a tool that does X” into something that actually exists — that layer is dissolving. Fast.
For businesses, this reframes everything. The question used to be: “Do we have the development resource?” Now it’s: “Do we know what we want? And can we tell if we got it?”
That’s a profoundly different question. And it favours different people.
Who Wins in a World of Clay?
If software is clay, then the valuable skill isn’t forming the clay — the AI handles that increasingly well. The valuable skill is knowing what to make.
This is great news for people with domain expertise. The operations manager who’s spent fifteen years understanding logistics workflows. The marketing director who knows exactly what their customer journey should feel like. The founder who can see the gap in the market but couldn’t previously afford to build the product to fill it.
These people were never short of ideas. They were short of a medium. Now they have one.
But — and this is the important caveat — clay doesn’t shape itself into something beautiful just because you sit at the wheel. The potter’s wheel metaphor cuts both ways. Yes, the material is more accessible. But taste, judgment, and experience still matter enormously.
Where It Breaks Down
Here’s where we get honest.
AI is extraordinary at implementing patterns it’s seen before. It’s remarkable at following specifications. It’s genuinely impressive at producing functional, working software from a description.
What it can’t do — not yet, and possibly not for a long time — is taste.
It can’t tell you whether your brand feels right. It can’t judge whether a user experience is delightful or merely adequate. It can’t decide whether your product positioning resonates emotionally with your audience. It can’t distinguish between a design that’s technically correct and one that makes people want to come back.
Subjective quality — the kind that separates good businesses from great ones — remains stubbornly human. Brand, voice, aesthetic judgment, the instinct for what will land with a particular audience: these aren’t specification problems. They’re taste problems. And taste is the one thing you can’t spec.
This is why the smartest businesses aren’t replacing their experienced people with AI. They’re giving their experienced people AI tools and watching what happens. The combination of deep domain knowledge, good taste, and AI-speed implementation is extraordinarily powerful.
The New Skill Stack
So what does this mean practically?
It means the skill stack is changing. “Can you code?” matters less. “Can you specify what good looks like?” matters more. “Can you review what was built and know whether it’s right?” matters most of all.
For businesses, this is an invitation to think differently about talent, about process, and about what’s possible. The people who understand your customers, your market, and your standards – those people just became more valuable, not less. Because now they can build.
Software is clay now. The wheel is spinning. The question is: do you know what you want to make?
—
Versantus helps UK businesses build with AI – not just talk about it. If you’re ready to start shaping, get in touch.