Job applications have become cheap to write and expensive to read, and both sides of the hiring market now use AI to cope with the other side's AI. But under Databanking, a vacancy could be matched against attested records inside the custodial boundary, and a suitable match could then arrive in the Owner's account as a faithful offer.
Applying for a job has become as easy as online shopping. AI writes a tailored résumé and cover letter in seconds, browser extensions fill in the forms, and application services promise ten times as many applications for less effort than one written by hand. A single vacancy can draw two thousand applicants in a day, of whom recruiters reckon only a small fraction are serious.
Recruiters have answered with automation of their own, first to filter résumés and now to conduct first-round interviews by voice. Candidates have answered in kind, and the logical end point has already been reached: one AI agent interviewing another, each politely promising that someone will be in touch.
The underlying difficulty is that everything crossing between candidate and employer is unverifiable text. Neither side can trust what it receives, and the cheaper that text is to produce, the less it is worth. Some recruiters now ask for friction to be put back into the process, which filters for persistence rather than suitability.
A Databank account is built from attested records rather than from self-description. Employment history, with dates, employers and pay, can be deposited from payroll. Qualifications are deposited by the bodies that awarded them. Home and work addresses are held for other services already. Together these describe a working life more accurately than any résumé, and none of it can be embellished by the Owner or generated by a model.
A Databank never analyses what it holds. It can, however, compete for Owners by offering bundled third-party analytics: services provided by independent, attested Requestors, run as approved computations inside the custodial boundary, with results placed back in the Owner's account. Job matching could be one such service. An Owner who opts in allows recruiters' vacancies to be matched against the record, and suitable job offers arrive in the Owner's inbox.
The recruiter, as Requestor, submits the vacancy as a registered matching computation. It runs as a standing, cross-sectional query over the accounts of Owners who have opted into job matching, and is evaluated again as new records are deposited. Once the position is filled, the query is retired and its offers are silently withdrawn.
The recruiter learns how many people match, which is useful in itself: a vacancy with no matches is a mis-specified vacancy, discovered before anyone has applied. Who the matches are stays inside the boundary. The Owner reads the offer and decides whether to respond, and only a response, released with the Owner's consent, puts the candidate in touch with the employer. Opting into matching and responding to a particular offer are separate decisions.
Some of the most useful inputs to a match are ones a recruiter cannot see, and in some cases may not ask about. Pay is the clearest example. A growing number of US states prohibit employers from asking about salary history, yet the gap between a person's qualifications and their current pay is one of the strongest indications that a better-paid role would be welcome. Inside the sandbox the matching computation can use that gap, and the employer never learns the figure. Addresses work the same way: a vacancy that would shorten the Owner's commute can be ranked higher without the employer learning where the Owner lives.
An offer that reaches an account in this way is faithful on both sides. The candidate meets the criteria in fact, because the criteria were evaluated against attested records. The employer is an attested and audited Requestor, so a scam posing as a vacancy cannot reach an account at all, and a vacancy that does not exist has a name and an audit trail attached to it. And each person is one candidate: the service-specific pseudonymous identifier that keeps a banned user from returning under a new account equally keeps one person from appearing as a hundred applicants. An invented applicant finds nothing to match, because there is no record behind him.
Candidates often pay for visibility today, through premium subscriptions to the platforms employers search. Under Databanking the recruiter pays for the matching computation, as any Requestor does, and a share of that revenue goes to the Owners whose records were queried. The screening that currently consumes recruiters' working days is done before anyone applies, so the fee buys a shortlist.
Matching depends on what the Owner has deposited. The Databank advises on account structure and on the services each arrangement of records makes available, and an offer is made only where the deposited data supports the match the vacancy requires.
Responsibility for lawful, non-discriminatory criteria stays where employment law already places it, with the recruiter. The evidence improves: a registered matching computation is on record, so the criteria actually applied to a vacancy can be inspected after the fact.
| Today | Databanking |
|---|---|
| Candidate describes themselves in a résumé | Attested employment, education and pay records |
| Applications generated and submitted in bulk | Owner opts into matching; one person, one candidacy |
| Recruiter screens thousands of applications | Vacancy runs as a registered computation over opted-in accounts |
| Employer asks about, or guesses, current pay | Pay used inside the sandbox, never disclosed |
| Candidate pays for visibility | Recruiter pays; revenue shared with the Owner |
Job recruitment is one instance of a general service: a third party matches opportunities against an Owner's record and delivers the result to the Owner's account. A clinical trial can find eligible participants among patients who have opted in to hear about trials. A property search can alert an Owner when a house matching a stated requirement comes on the market. In each case the opportunity travels to the data, and the Owner decides what happens next.