Anyone Can Produce Software Now? The Demo Gap

Engraving of a figure on a bridge, illustrating anyone can produce software debate

The claim that anyone can produce software now is spreading fast, and it’s strongest among people buying software rather than building it. Prompt an AI tool, wait a minute, and a working app appears. A closer look at what the tools’ own makers say shows a gap between a demo and a product that strangers rely on.

That gap is real, and it’s hard to see from below.

What the AI tool builders say

Boris Cherny, who created Claude Code at Anthropic, said on a podcast in February that he imagines a world a few years out where building software is open to everyone at any time. Andrej Karpathy, who coined the term vibe coding, said at a Sequoia event in April that everyone can vibe code anything.

Both also describe the part of the job that remains. Cherny says he hasn’t edited a line of code by hand since November 2025. Claude writes all of his code, at twenty or thirty pull requests a day. He still reads every line, and a human approves every change, because many people run the program. He puts the reason plainly: the code has to be correct.

An engraving after Munch's The Scream: a figure on a bridge with both hands to its face and its mouth open, under a sky of striped bands.
Image: thecoder.io

Why a demo hides the problem

A home-cooked dinner and a restaurant meal can taste alike on the night. Software is harder to judge. A fragile prototype and a sturdy one can look the same on screen and respond the same way to a click.

The difference appears once real users arrive. They click in unexpected sequences, use untested phones, enter odd data, and a few go looking for weaknesses on purpose.

Karpathy learned this firsthand. Months after he coined vibe coding, he built a real web app with it. The prototype came together quickly, but sign-ups, API keys, a domain name, payments and hosting ate the rest of the project, and he described stitching it all together as assembling a monster. His verdict: don’t use the method for anything that matters.

A hand-coloured engraving of a single honeybee with its wings spread, and beside it a piece of honeycomb with sealed cells.
Image: thecoder.io

Three kinds of “anyone”

The phrase anyone can produce software covers at least three people, and the claim means something different for each.

The first is a hobbyist who asks an AI for a small program for personal use. They’re the sole user and know its quirks, and a breakdown inconveniences nobody else. Karpathy’s original post described this case as acceptable for throwaway weekend projects, and Cherny takes the same view of his side projects.

The second is a technical person outside software engineering, such as a sysadmin or data analyst. Their apps can include logins and backups. Trouble comes later: an unmaintained library, a database query that crawls once tables reach a million rows, or a defect in a client’s one browser. Educator Matt Pocock says strategic mistakes can take nine months to come back at you. People change jobs faster than that, so the lesson never lands.

The third is a developer who has spent years watching software break in public. Typing speed isn’t their edge, and AI may do most of it. Experience means the next failure rarely catches them off guard.

A hand-coloured engraving of the same oak at three ages: a seedling with a few leaves, a young tree, and an old tree with a thick trunk and a wide crown, roots shown below each.
Image: thecoder.io

The illusion of understanding

Why do buyers miss this? A 2002 Yale study on the illusion of explanatory depth offers a clue. People rated their grasp of a zip, a flush toilet or a bicycle, wrote out how each works, then rated again. The second score was always lower. A demo gives the same false comfort: you’ve watched it work, but nobody made you explain what happens when two people save at once.

Vibe coding versus agentic development

People often confuse two terms. Vibe coding means prompting without understanding how software works, so there’s no way to judge the result. Agentic development means a developer steers the AI, reviews its output and knows where to be suspicious.

Simon Willison drew the line in March last year: if a model wrote every line but you reviewed, tested and understood all of it, that isn’t vibe coding. A year after coining the term, Karpathy retired it for professional work in favor of an approach with more oversight and no compromise on quality.

What “finished” requires

If anyone can produce software, the harder question is what finished means. A finished product has to serve people who had no hand in building it. It must run on devices nobody tried, handle data nobody predicted, keep users’ information apart, and stay secure through each framework update. When it fails, a named person has to answer for it.

Where developers fit next

Pocock borrows John Ousterhout’s split between tactical programming, which gets today’s feature working, and strategic programming, which keeps a system easy to change next year. He argues the models have largely taken over the tactical part. The strategic part still needs a person. Karpathy, after a month of letting agents write most of his code, said they make the subtle conceptual errors of a hasty junior developer, and that for code he cares about he watches them like a hawk.

Karpathy said vibe coding raises the floor, while professional work is about the ceiling, and that you remain responsible for your software. Sam Harris said on his podcast a year after a conversation about AI programmers that those left standing will be people with good taste who can point the machines somewhere sensible.

The practical test for buyers is simple: ask who can tell a demo that will hold from one that won’t, and who will be around when it breaks.

Source: thecoder.io

Written by

Saksham Bhargava

Saksham Bhargava runs Alteroid and has been writing about consumer technology for over five years, covering iPhone, Android, Windows, AI tools and gaming. With an MCA and six years of hands-on web development experience, he tests the features he writes about on his own devices and focuses on clear, practical guides that solve real problems. Away from the screen, he enjoys travelling to places most tourists skip.