Two experiences define job searching and hiring right now, and they're the same experience from opposite ends. Send fifty applications, hear back from two. Post one opening, get five hundred applications, and still can't tell which three are real. Candidates use AI to apply at volume because no single application feels worth the effort. Employers use AI to screen at volume because five hundred resumes can't be read by a person. Each side is reacting to what the other side just did, and the loop tightens every time it runs.

Blame the tools first, because that's the obvious target. Harvard Business School's 2021 "Hidden Workers" report found that 99% of Fortune 500 companies run an applicant tracking system, and 88% of employers admit that system filters out qualified candidates over exact keyword mismatches. Nearly half run a blanket filter that auto-rejects any resume gap over six months, whatever the reason. The report puts a number on it: roughly 27 million people in the US, employable and motivated, invisible to an algorithm for reasons that have nothing to do with whether they could do the job. LinkedIn has its own version of this. Recruiters who've used the platform for years say well-built Boolean searches return worse results than they did five years ago, and it isn't user error, it's a change in the platform itself. LinkedIn is paid in engagement and seat renewals, not in successful matches. A search that resolves in one try is worth less to the business than a search that takes forty pages.

But the tools are downstream of an older question. Why does a company exist at all.

Ronald Coase answered that in 1937, in an essay short enough to read in twenty minutes, and it eventually won him a Nobel Prize. Firms exist because the open market has costs. Finding the right person, negotiating terms, monitoring the work, all of that is expensive to do fresh every time. A company internalizes that cost by employing people directly instead of re-contracting for everything. Coordinate inside the firm cheaper than the market can, and the firm grows. Let the market get cheaper than internal coordination, and the firm should, in theory, shrink.

That theory is having a moment again because AI is supposed to be the thing that finally makes market coordination cheap. An agent can source, negotiate, and monitor a contract close to the way a manager used to. One piece I read called this the "Coasean Singularity," the point where company stops meaning a permanent entity and starts meaning a temporary coordination pattern, and career stops meaning a job and starts meaning a portfolio.

Satya Nadella has made the counterargument publicly: firms survive by encoding their own tacit knowledge into a private AI layer, which becomes the new source of cheaper internal coordination, permanently. It's a clean argument, and it also happens to describe the exact product Microsoft sells. That doesn't make him wrong. I'm just not going to treat it as a neutral forecast instead of a pitch.

There used to be a deal. William Whyte wrote about it in 1956, in a book called The Organization Man: you gave the company your career, and it gave you a career back, decades of it, in exchange for loyalty and the knowledge you accumulated along the way. That deal ended long before anyone was talking about AI. I watched pieces of it end directly, across eight years recruiting, including time inside Meta and Okta. SQL administrators. DevOps engineers. Knowledge managers, whose entire job was preserving what a company knew. All of it got rolled into other roles, handed to vendors, or automated somewhere along the way. The story everyone tells is efficiency. I don't buy it. A company doesn't route around its own memory because memory is expensive. It routes around memory because a person who holds it has leverage, and leverage is the one thing a company will pay a real cost to avoid handing anyone.

Same mechanism, different market. Research on pay transparency laws has found that employers accept measurably lower average wages, a real cost, specifically to avoid the renegotiation power that comes from workers seeing what everyone else makes. Companies don't optimize for cheap. They optimize for nobody having anything to hold them to. Knowledge management got vendored for the same reason wages compress under transparency laws. Not because the thing being suppressed was worthless. Because it was leverage.

So Nadella's version of the future has the mechanism backwards. Encoding tacit knowledge into a corporate AI model isn't firms discovering a new reason to value what employees know. It's firms finishing something that started the day the organization man deal broke: extracting the value of institutional knowledge without ever again depending on a person who could walk out the door with it. Nothing personal. Just the last item on a list that's been open since around 1980.

Which future wins, fragmented or bundled, comes down to a question that has nothing to do with how capable the AI gets. It comes down to who owns the layer that lets a stranger verify what you've actually done. Right now that layer is owned by platforms with every reason to keep it slow and unreliable, because slow and unreliable is the business model. There are real efforts to build something else. The Velocity Network Foundation is building open, portable infrastructure for verified work history that no single company controls. The EU's eIDAS 2.0 wallet is doing something similar for national identity. Whether either one wins against the platforms holding the keys right now isn't a technology question. It's the question capitalism asks of every open layer eventually: does it stay open, or does someone capture it, the way the internet did.

I'd like the open version to win. Wanting it isn't the same as trusting it. A fully verifiable, permanent record of everything you've ever done solves the trust problem and creates a new one nobody's talking about yet: the slack in a messy, self-reported resume disappears with it. A gap gets explained away right now. A bad year under a bad manager fades. A permanent, instantly checkable record doesn't forget, and whoever holds that layer holds a legibility project, not just a trust project. Legibility has always served whoever's doing the looking more than whoever's being looked at.

The resume, the profile, the record, none of it was ever really about capturing what someone can do. It was a proxy fight over who gets to hold leverage over whom. Employer and employee. Platform and recruiter. Verifier and verified. AI doesn't end that fight. It hands both sides sharper tools and moves the fight to a layer most people aren't watching yet. Whoever wins it decides what employer and employee mean next.