There are two stories about AI and software, and both are being told loudly to business owners right now.
Story one, from the hype merchants: development is over. Describe your app to an AI, and it appears. Anyone quoting you real money for a build is a dinosaur billing you for extinction.
Story two, from the frightened corners of the industry: AI code is a toy, sloppy, insecure, unmaintainable. Serious businesses should pretend the whole thing isn't happening.
We build production software with AI tools, real systems, with real users and real money moving through them. Which means we're positioned to tell you the third story: the true one, which is more interesting than either sales pitch. Here's what actually happens when AI builds half your app, and which half it can't touch.
What genuinely changed (and it's dramatic)
Let's give the revolution its due, because it's real.
The first version arrives absurdly fast. Work that consumed weeks, screens, forms, standard flows, the connective tissue between a database and a browser, now takes days, sometimes hours. Modern AI-assisted stacks (we build on tools like Lovable with Supabase underneath) genuinely collapse the distance between "described" and "clickable." This isn't marketing; Stack Overflow's annual developer survey shows the overwhelming majority of working developers now have AI in their daily workflow. The profession already voted.
Iteration became nearly free. The old economics punished changing your mind, every "actually, could we..." cost days. Now variations cost minutes, which quietly transforms product decisions: you can try three approaches and keep the best instead of committing blind to one. For clients, this is the biggest hidden win, the cost of exploring dropped through the floor.
The standard parts stopped being billable mysteries. Login screens, password resets, payment forms, admin tables, solved problems that every project used to re-pay for. AI produces them competently on demand, which is exactly why what apps cost to build is genuinely shifting downward for well-scoped projects. When a quote drops, this is the legitimate reason it drops.
What didn't change (and this is where projects still live or die)
Now the half the hype leaves out.
AI builds what you describe, including your mistakes, at speed. The tools have no opinion about whether your app should exist, whether the workflow matches how your staff actually operate, or whether the feature list traces to a real problem. Describe the wrong product and AI delivers it faster than ever, wrong at unprecedented velocity. Every cause we documented in why software projects actually fail survives the AI era untouched, because none of them were ever coding problems. The expensive failures were always decision failures, and the deciding is still entirely human.
The invisible layer still needs adult supervision. AI-generated code is competent and confident, including when it's wrong. The places it goes wrong cluster exactly where stakes are highest: security rules, payment failure states, data permissions, what happens when the network drops mid-transaction. The code runs; the demo shines; the hole waits for a real user. This is why "AI built it in a weekend" apps demo beautifully and die in production, nobody reviewed the half that doesn't show. Our working rule: AI drafts, humans audit, and the audit concentrates precisely where the demo can't reach.
Maintenance inherited everything. AI-built software is still software, dependencies age, platforms shift, the living-thing economics we've written about apply in full. If anything, review matters more over time: a system assembled fast needs its documentation and structure kept deliberately human-legible, or year-two changes become archaeology again.
The honest reframe: what you're actually paying for now
Here's how the arithmetic of a software project has genuinely reorganized, and it's worth stating plainly because it changes what you should buy:
The typing got cheap. The judgment didn't move.
Scoping the right product, designing the invisible layer, auditing what AI produced, deciding what not to build, keeping the system alive, that's now the substance of the work, and of any honest quote. When we build with AI tools (openly, it's in the article title), clients aren't paying us to type what a machine could type. They're paying for the thousand decisions the machine executes but cannot make, and for standing behind the result with our name on it.
So the practical guidance, whichever builder you use:
- Distrust both extreme stories. "AI makes it free" and "AI code is garbage" are both selling something. The truth is a powerful tool wielded well or badly.
- Ask any builder how they use AI and what they review. The strong answer names both, acceleration and audit. Silence on either end is a flag.
- Expect the savings in the right place. Standard parts cheaper, iteration cheaper, scoping, security, and judgment priced like the load-bearing work they now visibly are.
- Your homework didn't shrink. The problem definition, the one-sentence MVP, the success metric, AI made everything downstream of clarity faster, which makes clarity itself more valuable, not less.
The tools changed the speed of building. They didn't change what building is for, and the businesses that understand the difference are about to get better software, faster, than any generation before them.
The ones that don't are about to get the wrong software at record pace.