The First Move

This essay is my read on how that happened and what I think engineers should do about it. I have sat on several sides of the job post. I started as an engineer in 2010. As a team lead at past employers I was the hiring manager for hundreds of roles. I have interviewed more than 2,000 engineers, and since 2023 I have been building TeamEx. The last ten engineers I hired for TeamEx, I found by searching. The founders and hiring managers I talk to are moving the same way, searching more and posting less, while the engineers I talk to keep telling me about job applications that never get a response. Both groups are looking at one shift from opposite sides.

For thirty years, the job post was the unit of the job market

MonsterBoard went online in 1994 and CareerBuilder followed in 1995.[1] Everything built since has assumed the same unit of work, which is the job post. A company writes one, engineers apply, somebody filters the pile, a few people get interviews and one gets hired. The engineer moves first, and the company chooses from whoever showed up. Thousands of job boards grew up around the post. LinkedIn won by becoming the best place to publish one and the best place to answer one. Applicant tracking systems exist to manage what a post brings in, and internal hiring pipelines are drawn as funnels that start with it.

It worked for a long time. I hired that way myself for years, as the hiring manager for my own teams. When a role could be described by a stack and a number of years, a post was a fair way to say what you wanted, and an application was a fair way to answer.

The two companies that started it did not make it to the end. Monster and CareerBuilder merged in 2024 and filed for Chapter 11 in June 2025.[1] Their bankruptcy is a convenient date stamp for the end of what I call the job-post era.

The role outgrew the job description

A classic job description lists technologies, years of experience and responsibilities. That format assumed the engineer’s job was to write code in a named stack. By the spring of 2026, about six in ten people who build software were using AI agents at work, roughly double the year before.[2] When an agent writes much of the code, the engineer’s job moves to the work around it. They frame the problem, do the research, direct and check the agents, and decide what is good enough to ship. I described that split in my essay on founder-engineer fit, where the first 20% and the last 10% of a task are the parts a person still owns.

So the companies I work with have stopped asking for a JavaScript engineer with 15 years of experience. They ask for system thinkers and analytical thinkers, people who can run research and manage agents. You can type those words into a job description, but they don’t filter anything. Every applicant will claim them, and no job board has a checkbox for how someone makes decisions.

Job descriptions are also written for keyword match, because job boards show a post to people whose profiles carry the same keywords. That worked when the keywords were the job. A post built around “JavaScript” and “5+ years” still reaches the people who match those words, and these days it tells neither side anything about the qualities the company is hiring for.

AI writes the post, AI writes the application, and AI reads both

Then AI arrived on both sides at once. Hiring managers generate the job description and, in my experience, often publish it without reading it closely. Applicants use tools that rewrite the resume for every post, aimed at the keywords in it, and some of those tools send the applications too. One post now brings in more than two hundred applications on average, about twice as many as three years ago.[3]

The pile lands in an applicant tracking system, where more AI shortlists it. So an AI-written application answers an AI-written post and gets ranked by a third AI. Nobody in that chain has to read anything, and often nobody does.

What this loop selects for is the quality of the applicant’s AI. The person who gets the interview is the one whose tool wrote the closest match to the post, and the engineer who could do the job best can sit deep in the pile, unopened.

A hiring manager described it on Reddit in March 2026 better than I can. One post brought 400 applicants in a few days, and almost every resume had been tuned to the job description by an LLM. When the manager used an LLM to sort the stack, “It pulls the most gamified resumes to the top of the stack.”[4]

This is why so many qualified engineers send application after application and hear nothing. The engineers I talk to describe the same thing. They fit the role, they apply, and no person ever looks. One engineer in San Francisco applied to about 670 roles and sent more than 1,000 messages without one concrete opportunity.[5]

Job posts turned into marketing

The post itself also stopped meaning that a job exists. By the best counts I have seen, about one in five job posts is for a job nobody intends to fill, and four in ten hiring managers say their company posted a fake listing in the past year.[6] Companies post to look like they are growing, to collect resumes for later, or to reassure an overworked team that help is coming. Staffing agencies, in my experience, post roles they do not have so that their database of resumes keeps filling. It has become common enough that New York’s legislature passed a bill in June 2026 that would make larger employers state in every ad whether the position is a current vacancy.[7]

For an engineer, it means some share of the applications you send go to a job that was never there.

So the first move changed sides

The role no longer fits a description, the pile is written and read by AI, and many posts are not real jobs. Put those together and the job post is dead as a way to hire, for the company as much as for the engineer. The need is specific, the inbound pile is mostly noise, and reading it costs more than it returns. So the first move is changing sides. Companies go out and find the person, and the post, if there is one at all, is a formality.

In March 2026 a how-to post showing an AI tool applying to 50 jobs in under 30 minutes went viral on X. Gergely Orosz, who writes The Pragmatic Engineer, answered it with “The death of inbound applications is upon us.”[8] The hiring numbers already agree with him. People a company goes out and finds are a tiny share of its pipeline and are nearly eight times more likely to be hired than people who apply.[9] At the smallest startups, about three in ten hires are now found that way.[10] Hiring managers say in public that they no longer take inbound seriously, and some small companies have given up on it after too many AI-written applications.[11] I see it in my own work. The founders and hiring managers I talk to are cutting job posts and searching instead, and I did the same for my last ten hires.

AI made the mess on the inbound side, and it is also what made outbound cheap. Searching used to mean Boolean strings and a sourcer’s time. Now a founder describes the person in a sentence, and there are many AI tools that will run that search.

LinkedIn, the giant of the job-post era, is still built around the post and the self-written profile. Its search filters on what members typed about themselves, and it has to fight fake accounts at a scale that is hard to picture, with more than 110 million stopped or removed in the second half of 2025 alone.[12] I went through the difference in detail in how LinkedIn search compares with search on evidence.

You are being indexed whether you take part or not

Search runs on data. Sourcing tools collect what is public about you, starting with your LinkedIn profile and extending to GitHub commit history, Stack Overflow answers, open-source work and conference talks.[13] They build a profile of you from it, index it, and make it searchable by people you have never heard of. Nobody asked you. You probably exist in ten or more of these databases right now, and you maintain none of them, so you don’t control what gets found or how you are presented.

That makes the job search look a lot like search engine optimization. A website that wants traffic makes sure Google can read it. An engineer who wants to be found and approached with the right opportunities has to make sure the tools doing the searching know as much true information about them as possible. I call this findability, and I think it is now the engineer’s responsibility in the way a resume used to be.

One engineer with eight years of experience sent 130 carefully targeted applications and did not get one phone screen. Every interview in that two-month search, including the one that became an offer, came from recruiters who messaged first.[14]

A great resume and a tool that applies for you do not help with this, because fewer and fewer people are reading applications. A LinkedIn profile alone does not either, because everybody has one.

What you say is cheap. What your work says is what gets you ranked.

There are two kinds of data about you. The first is what you say you are, let’s call it the claim layer: your headline, your skills list, the summary at the top of your resume. You can fix it this afternoon. Update your LinkedIn profile and the tools pick up the change the next time they refresh.

Do that. But everybody else can do it this afternoon too, and with AI most of them will. Once every profile says “systems thinker who ships with agents,” the phrase stops separating anyone.

The second kind is what your work shows and says about you — the proof layer. It is what you built and where, the code and the pull requests, how you framed a problem for an agent and what you did when it went wrong, a talk, a write-up, a public argument about a design choice. This is the data that tells a company how you work, think and communicate, how you make decisions and how you lead. The role now requires exactly those things, and a job description could never filter for them. Proof takes years to accumulate and cannot be produced in an afternoon, so it keeps its value after everyone else has polished their claims.

The engineer who sent 670 applications is one illustration. He stopped playing for volume and spent his time building products, publishing and meeting people, and by his own follow-up more than 80 recruiters then reached out to him between January and May.[5] Jason Fried gave another in September 2026, when he announced that 37signals had hired Rob Zolkos. Zolkos never applied, and no role was open. He had published a write-up on one of their products, a flipbook built from its GitHub commits and a command-line tool for its API, and an employer noticed.[15]

What counts as proof of work

Proof of work is evidence of what you did, tied to a source and a date, that someone else can read. Code is the obvious kind, meaning repositories, pull requests and review comments. Coding-agent sessions are the newest kind and, I think, the most telling, because the log of how you directed an agent is a direct record of the job as it now exists. Talks, streams and published writing show how you explain things. Posts show what you pay attention to and what you think about it. A recorded interview on one area shows how you reason when someone pushes back.

The trouble with proof of work is that it is scattered. It sits in a repository, a session log, a video and a thread, no search can filter on it in that state, and nobody has time to keep a profile current on top of doing the work. So imagine one place that you connect, once, to the places where you already work and talk: your coding-agent sessions, your GitHub, your blog, the accounts where you post. You choose what to connect. From then on it follows along. It gathers the evidence from the work you already do every day, with no extra effort from you, and turns it into a structured profile of what you built, the skills you used, and how you work, think and communicate, with every item tied to its source and date. You never update it, because your work does. When a company searches, it finds the current you, on a profile you control.

Two cautions. Most engineers’ best work is private, inside a company repository, so public code alone undersells a lot of very good people. That is why session logs and interviews matter as much as GitHub. And dead does not mean gone. Posts still go up, and inbound is still the largest single source of startup hires, at about four in ten, though it takes roughly 300 or more applications to produce one of them.[10] I would keep applying to the few roles you really want and spend the rest of the time on being findable.

How I built TeamEx around this

I run a company built on this view, and TeamEx is the place I just described. An engineer signs up and connects their work: coding-agent sessions, GitHub, Hugging Face, YouTube or Twitch, writing, X and LinkedIn. The session connector uploads the agent logs and never the code. Then you live your normal life, and the profile stays current as you work. You can add AI interviews, or interviews with expert interviewers, to go further. The way profiles are built is written up in full on the site.

Founders search those profiles in plain English. Results are ordered first by how well a profile matches the search, then by profile strength and how recent the evidence is. Your identity is protected until you accept an interview request, with no name, no photo and no current employer shown, so your current employer sees nothing. I think it is the best way to stay in the market and stay open to new opportunities without announcing it. Joining, connecting your work and getting interview requests cost you nothing.

I think the next ten to twenty years of the job search will be organized around this. The job post will survive as paperwork, and the hiring will happen in search. If you hire, the first move is yours now. If you are an engineer, it no longer is, so fix your claim layer and focus on building the proof.

Don’t look for a job. Let founders approach you.