America is importing the very workers it is now laying off.

The people who wrote this build software for a living: production systems, security audits, the interfaces most users never think about. That makes us the workers the H-1B program now overwhelmingly brings in, which is the reason we have something to add to a debate usually conducted without us.

Each of us now works with an AI model as a matter of routine. The tool differs by team, whether Claude, GitHub Copilot, Cursor, Codex, or ChatGPT, but the effect does not. A task that used to take a mid-level engineer a day now starts with a few sentences of instruction and a draft that arrives in minutes; the work is to check it, correct it, and decide what to keep. Most of the people who write the country’s immigration law have never done this job. We do it every day.

Every year, a computer at U.S. Citizenship and Immigration Services runs a lottery that decides which H-1B visas are approved. It doesn't measure skill or need. It draws names out of a pool of hundreds of thousands, and the people it draws win the right to work in the United States for the next three to six years. This is how the country fills many of its most sought-after engineering jobs.

The program is almost always described by a single number: 85,000, the visas the lottery hands out each year. That figure captures only the capped part of it. Universities and nonprofits hire beyond the cap with no limit of their own, and once they are counted, the government approved roughly 141,000 people for new H-1B jobs in 2024. Because each visa can last up to six years, the number of people actually working on one at any given moment is far larger. The only time the government counted the full standing population, in 2019, it reached 583,000, and every estimate since places it above 600,000.

No one setting out to recruit talent would build the system this way. Seven in ten recipients come from a single country, India. That is the sign of a channel narrow enough to be worked, not a judgment about the workers in it. Its heaviest users are staffing and outsourcing firms that hold the visas and place the workers at other companies for a fee.

The program selects for none of the things a talent system should select for: not skill, not scarcity, not pay. It selects for the firms best at filing the paperwork.

For thirty years the case for importing engineers rested on one premise: the work took more skilled hands than the country could supply. Generative AI has ended that premise. At the central task, writing code, the best models now outperform even a top software engineer. They produce more of it, faster, and with fewer bugs, and they improve every few months. Work that once took a team of programmers now takes one person directing a model. The number of engineers a company needs to ship the same software has dropped, and it keeps dropping.

There is a second effect, and it cuts against the program’s founding premise. The work is shifting from writing code to describing, in plain English, what the code should do. As the work changes, the skill that matters changes with it. The engineer who can state a problem precisely, catch the machine when its answer is wrong, and explain the tradeoffs to the people around him is worth more than the one who can only write syntax. The job now rewards judgment and clear communication, and neither is a capacity the country needs to import.

The executives building these tools have said as much. In a June 2025 memo, Amazon’s chief executive, Andy Jassy, told employees that AI would “reduce our total corporate workforce” in the years ahead. Amazon files more H-1B petitions than any other company in the country. In its own words, it is planning to need fewer of them.

The usual objection to criticizing the program is that it sounds like a complaint about foreigners taking jobs. The stronger objection is to what the program does to the people inside it, on both sides. An American engineer competes against a worker the same employer is permitted to pay less. The visa holder is tied to one company by a green-card wait that can run past his working life; he cannot easily change jobs, has little leverage to ask for a raise, and can be sent home within weeks of a layoff.

A worker who cannot leave has little leverage, and the wage he accepts sets a floor under the American working next to him. The firm that holds the visa keeps the difference. Both workers lose; the intermediary does not. A change that helped only one of them would miss most of what is wrong here.

None of the changes below would close the country to skilled immigrants. Each one removes a specific opening that employers and staffing firms have learned to work: the lottery that let a single worker be entered under a dozen names, the wage tiers that price most jobs below the local median, the green-card backlog that keeps a worker from ever leaving the company that sponsored him. Fix the openings and the program shrinks on its own.