Two AIs Fought Over Who Gets to Apply for My Job

I was applying for a product design role at Synthesia when my own software started fighting with itself. One AI had already researched the company, tailored my resume, written a cover letter, exported the PDFs, checked the filenames and opened the application form. It was treating this like a job I actually wanted.

Then Tsenta appeared on top of the page. It detected the Ashby form, opened its own control panel and offered to generate another cover letter, choose a resume strategy and submit the application. “Review before submit” was turned off.

The first AI was saying: this role matters, let’s get it right. The second AI was saying: bro, it’s a form. Move.

Two AIs were now fighting over who got to pretend to be me.

This is probably the clearest picture of where job hunting is going. We are already past the era where AI merely helps you write a cover letter. That now feels almost quaint. We are entering the era where several agents can occupy the same application at once, each with a different strategy, different documents and a different idea of how badly you want the job.

One AI sees Synthesia as a company I deliberately chose. The other sees an endpoint with text fields, upload buttons and a submit action. It does not care whether I have spent three hours researching the company or whether I discovered the role eleven seconds ago. To the agent, both situations look structurally identical.

The job application has become contested territory. One agent wants to make the application more specific. Another wants to make it more abundant. One is trying to explain why my strange combination of design, technology, research and founder experience makes sense for this role. The other is trying to increase throughput.

One says: this application is special. The other says: this is application number 438.

The annoying part is that both may be right. A deeply researched application can still be rejected because it lacks one keyword. A generic application submitted 1,500 times can accidentally reach exactly the right person. A mediocre application sent everywhere may produce more interviews than a beautiful application sent once.

A software engineer named Julian Joseph once ran an auto-application bot on two laptops while he slept. By morning, the machines had submitted close to 1,000 applications. Eventually they sent around 5,000 and produced roughly 20 interviews.

That sounds impressive until you compare it with his manual search. He had previously produced about the same number of interviews from only a few hundred carefully submitted applications. The machine was much worse at applying. It was simply much better at not getting tired.

And maybe that is enough. The bot could keep buying lottery tickets while Joseph spent his actual time networking. The automated applications created volume. His human relationships produced some of the exceptional opportunities, including interviews with Apple and the White House.

That split may become the default job-search strategy. Let the machine apply everywhere. Use the human for the places that matter.

The problem is that “everywhere” is also where everyone else’s machine is applying. Recruiters can publish a vacancy and receive hundreds of applications almost immediately. Many come from candidates in the wrong country, without the required clearance, without the necessary experience or without any realistic ability to accept the role.

The candidates may never have seen the job. Their agents saw it.

Everyone is behaving rationally. Candidates automate because applying manually takes forever and most companies never reply. Employers automate because they receive too many applications to inspect. The automated filters reject qualified people, so candidates respond by submitting even more applications.

Nobody designed this system. Everyone made the locally sensible decision, and together they created a completely cooked labour market.

The application itself is losing value because the cost of producing one is approaching zero. A tailored cover letter used to signal that someone had invested time. Now a model can read the company website, imitate its language, mention the correct product and generate a convincing declaration of lifelong passion in seconds.

A clean resume used to signal communication skills. Now every resume can be clean. A thoughtful answer to “Why do you want to work here?” used to suggest interest. Now an agent can produce 500 different explanations of deep personal interest before lunch.

AI has not made every applicant equally qualified. It has made every applicant equally capable of looking qualified for fourteen seconds.

This is where the fight between my two AIs becomes more than a funny browser-extension conflict. The tools represent two different theories of effort. The first believes AI should help me care more deeply about fewer applications. The second believes AI should allow me to care less about many more of them.

But the moment an agent starts applying for me, the question is no longer whether it can fill in a form. The question is whether I trust it to represent my life.

Would I actually take this job? Would I move to that country? Do I require sponsorship? Do I genuinely have the experience the application claims? Is the agent presenting me persuasively, or quietly inventing a more employable person?

The more autonomous the agent becomes, the less it feels like a tool. It starts feeling like delegation. I am giving software permission to represent my identity, ambitions and legal circumstances to companies I may not even know exist.

My agent can apply while I sleep. Their agent can reject me before I wake up. Another AI can then invite me to an interview, record my answers, score my personality and disappear without a human entering the process.

There are already applicants describing exactly that. An AI submitted the application. Another AI conducted the interview. Then nothing happened. The machines completed the hiring ritual without producing a job, a relationship or even a proper rejection.

It is tempting to say companies should ban AI applications, but that is not going to happen. Employers cannot expect candidates to spend thirty minutes manually copying a resume into Workday while using AI to reject the same application in three seconds. That asymmetry is too stupid to survive.

Job applications will become automated because much of the process should have been automated years ago. Nobody’s character is revealed by typing their address for the seventeenth time. The interesting question is what becomes valuable after the application itself loses meaning.

Probably inconvenience.

Not pointless inconvenience like creating another account and entering your employment history again. Meaningful inconvenience. Something expensive enough to prove that you actually care.

A project made specifically for the company. A contribution to its open-source repository. A conversation with someone on the team. A public body of work. A trial task. A trusted introduction. Showing up somewhere in person.

One young job seeker spent months applying online without getting replies. Eventually, he walked into a store carrying a paper resume and got a job that had not even been advertised online.

A paper resume now feels disruptive. That is hilarious.

After decades of digitising recruitment, the new premium signal may be a person entering a building and saying hello. When applying becomes free, showing up becomes valuable. When writing becomes free, taste becomes valuable. When every candidate can express perfect enthusiasm, evidence that you cared before the vacancy existed becomes valuable.

Maybe job applications eventually disappear. Your personal agent could maintain a verified model of your skills, projects, reputation, salary expectations and availability. A company agent could expose what the team actually needs. The two agents could compare fit and negotiate the basics before either human spends time on an interview.

That could be much better than what we have now. But the transition will probably be a mess.

We will install five extensions. One will rewrite the resume, one will generate the cover letter, one will fill the form, one will submit automatically and one will warn us that the others are making mistakes. They will all claim to know us. They will all want control of the Apply button.

And somewhere underneath the panels, pop-ups and automatically generated declarations of passion, there will still be a person trying to decide whether they actually want the job.

Connected reading

• What Kind of Worker Is AI Allowed to Become?
• The Browser Is Becoming the Agent's Body
• The Product Needs a Harness
• The Last Obstacle Was a Popup
• The Agent Becomes Real When It Can Pay

AI authorship note

This essay began when Kris Haamer opened a carefully prepared Synthesia application and watched Tsenta appear on top of the same form, ready to generate its own materials and submit without review. Kris recognized the scene as two competing theories of AI job hunting fighting over the Apply button. He pushed the piece away from a generic comparison of auto-application products, asked for human anecdotes from applicants and recruiters, insisted on his own voice and paragraph structure, and selected the final version for publication. AI supported product research, source gathering, drafting, editing and implementation.