Who really owns your product: AI makes irresponsibility cheaper upfront — and more expensive later

“Finally, I can build my product without developers.” 

Beautiful: no meetings, no estimates, no “this will take two sprints.” No senior engineer explaining why your “small change” is actually a database migration, three edge cases, and a future lawsuit. 

Just you, your genius idea, and an AI agent saying: “You are absolutely right. I shall update the product.” 

And then $50 disappears, then another $100, then another $500. 

And somewhere between “make it better” and “why is nothing working now,” you may realize something uncomfortable: you didn’t replace developers. You replaced a team with a slot machine that speaks fluent confidence. 

Those Damn Developers Won’t Tell Me What to Do 

Anthropic & Co. are moving from “AI as a tool” toward “AI as an agent that replaces a development team” — and, potentially, later, any team doing intellectual work. 

And yes, they are getting some genuinely interesting, genuinely impressive results. 

But here’s the funny part: the people who didn’t want to become “slaves to developers” — anyone who has ever been a developer is quietly dying at that wording — may simply become slaves to something else: tokens, APIs, models, infrastructure, and American corporations. 

Congratulations! You didn’t escape dependency - you upgraded to a more expensive cage. 

The Fake Software Company Inside the Prompt 

In case you’re not following what’s happening, here’s my pre-chewed recap - they are already building fake software companies inside prompts. 

There’s a Manager: “You are the project manager, your responsibilities are… you get as input… you provide…”. Basically, a compressed extract of what we think a manager is. 

There’s an Analyst: “You are a System Analyst and you read stakeholder inputs, analyze them, and provide documents like…” 

There’s a Team Lead (because apparently every little fiefdom needs a commander); a Developer, a Tester, An Architect (because of course there has to be one) - the whole corporate terrarium. 

Then they take a business idea and “run” it through this little role-playing theater, in a sequence that tries to emulate how things happen inside software companies. 

Business idea → AI agents discussing it with each other — a hallucinating committee in a trench coat — specifications → plan → tickets → code → tests → more code → more tests. And so on. 

Anthropic especially seems to enjoy this game. They are pushing it, showing it off, and honestly, they do have something to show. There are already stories about complex projects being rewritten or ported at absurd speed — millions of lines of code, timelines that would have sounded insane two years ago. 

Very impressive. But also, let’s ask the boring adult question: “What do you actually get?” 

The Prototype Trap 

At first, maybe your idea really does get implemented faster, maybe even better. 

You don’t need to find people, negotiate with them (and people can be different, you know) wait for estimates, pay salaries, manage moods, explain the same thing sixteen times. 

Compared to the old order of things (pre-AI prices, pre-AI timelines) it may feel like magic. 

But in many cases, what you get is still a prototype. Maybe a very shiny prototype, maybe even with blackjack and hookers. But still a prototype. 

And then comes the real question: “Is it a product you own?” 

And what does “own” even mean here? 

Legally, is it yours? Probably yes. 

Financially, do you pay for it? Absolutely. And you may be surprised how much. Very soon, “a developer’s salary” and “burning money on tokens” will become numbers you can actually compare. 

But operationally, do you own it? Can you run it, maintain it, debug it, improve it, refactor it, explain it, defend its architecture, recover it when something breaks? 

That’s where the problem begins. Because in theory, yes, you can do anything. 

In practice, no. 

A Boat Anchor With a Stripe Subscription 

Product polishing is not one heroic push, it is hundreds, sometimes thousands, of changes - tiny, annoying, contradictory. 

Changes you only discover after users touch the product and immediately behave like users, meaning: completely wrong, but somehow also correctly. 

Understanding what to improve and how to improve it takes a ridiculous amount of effort. 

The deeper you go down the rabbit hole, the more clearly you understand how far you are from the ideal. Or maybe not even the ideal. Maybe just from “minimally acceptable.” And I’m not sure AI saves you here. 

At some point, AI may stop being a tool and become a second voice that drowns out your own product judgment. 

Instead of clearly formulating what is wrong, you start typing: “Something feels off here. Make it better.” 

Great! Now the hallucinating committee is also in your head. 

And when you have a gazillion lines of code (by the way, a gazillion is not even that much in software; before AI, an average team could ship more than that in a single release) someone has to navigate it. 

The more code there is, the harder it gets. And sooner or later, you realize what you have built is not really a clean product. 

In my language, we’d call it a suitcase with no handle - too painful to carry, too expensive to throw away. 

In more native English: a boat anchor with a Stripe subscription. 

Because the model made it for you. And after some time or maybe from Day Zero you may not understand what is arranged where and why. 

You can no longer influence technical decisions because there is nobody in your boat who actually gets it. 

Nobody who controlled what was being created at every stage. Nobody who can say: “No, we don’t build it that way.” 

And that is the real difference between using AI as a tool and letting AI become the only “person” in the room who understands your product. 

Who Really Owns the Product? 

When the excitement over the genius of your own idea wears off and sobriety kicks in, you may look at your already implemented idea and realize how many flaws it has. And all those flaws were built at maximum burn and with zero mercy. Simply because: “Why not? We can.” 

People will say: “Everything AI created, AI can also fix.” 

Sure, as long as you still understand what needs to be fixed. And how. 

You will hear this many, many times: “You are absolutely right! I shall update your product to also support…” And many, many dozens of times, you will spend $50, $100, $200, $500 on yet another small improvement. 

Here’s the fun little nuance: the faster your product brain works, the faster you can burn money on tokens. In other words: the smart ones go broke first. 

Because the loop is addictive: idea → prompt → output → dopamine → flaw → prompt → more output → more flaws → more tokens. 

And then, when you finally understand how many changes need to be made and how much each round costs, I can’t guarantee it, of course, but there is a very high chance you will be stunned as hell. 

And then no developer will be able, or willing, to support it afterward. 

Because Brandolini’s law (the Bullshit Asymmetry Principle) was, of course, originally about something slightly different, but it fits here 100%: “The amount of energy needed to refute bullshit is an order of magnitude bigger than to produce it.” 

Generating is easy. Cleaning up, understanding and ownership are expensive. 

A quick note: the leading models that work reasonably well (as reasonably as possible today) in unsupervised mode can cost around $25 per million generated tokens for the previous generation, or around $50 for the latest one. Plus a pile of overhead. And 1 million tokens is VERY easy to burn through. 

In one day, you can burn tens of millions if your eyes are glowing with enthusiasm. 

So the most important question is not: “Can AI build my product?”. Increasingly, yes, maybe. 

The real question is: “Who actually owns the thing once it exists?” 

If you can’t fix even the simplest bug without spending X × $50, is it really your product? If nobody understands the architecture, is it really your product? If every improvement requires another round of token roulette, is it even a product? 

If you were not able to build a team that feels like it is in the same boat with you, then how is this different from a casino? And in a casino, by the way, someone does win from time to time.