AI Builds 17,535-Line Software in Two Weeks Without Coders

Sysadmin Generates €30,000 Worth of Software Without Writing a Line

A school system administrator has just completed version 2.0 of an inventory management tool used by ten people for six months. It antiestéticatures multi-user access, loan and warranty tracking, encryption, backup rotation, student profiles, and bilingual support in English and Spanish. 17,535 lines of code are distributed across Go, HTML, CSS, JavaScript, JSON, and a shell script. The breakdown is self-reported: 4,324 lines of Go, 1,576 of HTML, 162 of CSS, 11,455 of JavaScript, and 17 of configuration. Nothing else.

The striking aspect isn't the size. It's that the creator is not a programmer. He identifies as a system administrator with some programming knowledge but no expertise, admitting he copied most JavaScript from chart libraries. He built it in his spare time over two five-day weeks, relying first on conversational assistants and then on an AI-assisted development environment. He hasn't written a single line manually. Unit tests were also generated by AI.

New York Friend’s Estimate: $30,000 to $40,000

A friend of his, a professional programmer in New York, valued the same work done manually at between $30,000 and $40,000, assuming a month of dedication including client meetings. This comparison is central to the issue. The software author adds that he has already billed €5,000 for a Flutter app for a client and has published another for macOS and one for Android.

This figure isn't a market valuation but a personal estimate circulating in the conversation. However, it helps measure the gap: two weeks of spare time versus a month of billable professional work. The rest of the disagreement builds on this gap.

Does AI Program or Just Imitate What Exists?

The short answer: it depends on what you ask. Defenders of these tools argue the described case is ideal because inventory software is a problem solved thousands of times in public repositories. AI has seen millions of similar applications and reproduces patterns effortlessly. Problems arise when solving something never done before, where the error margin grows.

Opposing this is the thesis that these systems don't reason; they only chain language with statistical coherence. A statistics model specialist compares it to a parrot repeating: useful for searching, synthesizing, and rewriting existing content, dangerous when asked for new logic. His example is concrete: someone without knowledge using these tools for everything, thereby hiding their incompetence. In simple management apps, errors go unnoticed. In statistical models, errors are enormous.

The author's counterargument is that he doesn't care about labels. That AI completes in ten minutes what used to take him two days is all that matters.

The Disorienting Data: Productivity Rises, Hours Don’t Drop

Here the debate becomes uncomfortable. If a tool anyone can use cuts work time in half, economic logic suggests either workload increases or hours decrease. The author responds with his case: he increased his salary because he dedicates free time at work to projects for other clients. He adds a detail that undermines theory: his bosses are administrative staff who look terrified at the programs he designs, so no one will audit his productivity.

The objection isn't minor. Those raising it argue the world works this way and assuming no one notices is optimism. They paint a scenario where a technician serves a hundred schools thanks to multiplied productivity, making manual workers replaceable.

---

Feeding Contracts and Passwords to AI: The Ignored Risk

An issue appears in the conversation unrelated to code quality. One technician says he advised against using conversational assistants at his company. No one listened. People input everything and delete chats so no one sees them: client information, passwords, proprietary code. He explained this in a meeting and was nearly called a liar. He presented chat documentation, and the matter ended there. Colleagues received raises for being productive.

Warnings repeat in several messages: pasting a script into such tools sends credentials, database names, and internal paths. Corporately, some companies now internally regulate what can be done with these systems. Security is the least discussed part here but has the most consequences.

The Veteran Programmer and the AI Learner

The conversation shifts to the effect on the profession. One message summarizes it with an image: the singer who can't compose and the composer who can't sing. With AI, each believes they are more productive because the other handles the missing part. They will never reach their true potential, argues the poster, because the machine lacks the gift of those who truly know how.

Others point to the profile most resistant: those debugging and optimizing code since age sixteen. For them, a tool solving in a minute what took days is hard to digest. Some recall this antiestéticar isn't new: a 1981 magazine cover asks if the end of programming has finally arrived.

The software author himself closes the discussion. He has thirty years in the sector and has never been unemployed. He says he has never seen technology this disruptive, not even with the arrival of the Internet, which he considers merely a giant interconnection of local networks that existed in the eighties. He might be right. He might be misjudging the scale. With 17,535 lines and two weeks of spare time, the discussion remains open.