Adecco put a number on its agent rollout.
On August 6, Adecco Group chief executive Denis Machuel said the company had already hit its full-year target of 50 percent of revenues being agent-enabled. That is Adecco's term for revenue where an AI agent performs part of the work of filling the job.
He then raised the target to 70 percent coverage by the end of the year. That 50 percent covers the Adecco brand business rather than the group total, and the group runs two other units alongside it.
The supporting detail came with the results. Recruiting agents are live in 10 countries and have handled 2.2 million conversations.
Adecco credited the rollout with a 10 percent gain in fill rate, a 40 percent cut in time to submit candidates, and recruiter productivity gains of 25 to 35 percent. Those are the company's own figures, disclosed on its earnings call, and nobody outside the company has measured them.
The quarter around them was solid. Revenue rose 5.6 percent on an organic basis, adjusted for trading days, and EBITA rose 21 percent at a 2.8 percent margin. The shares still fell about 4 percent after the release.
Kelly, reporting the same day, described a technology overhaul with a unified AI-powered CRM platform that it said is helping sales and productivity. That is customer-relationship software rather than a recruiting agent rollout, and Kelly attached no coverage figure and no recruiting metrics to it.
Pedagogue Systems' view. A fill-rate gain and a margin gain landing in the same quarter does not show that one produced the other, and Adecco did not claim it did. How Adecco arrived at the claim is more instructive than its size.
Every operator trying to prove an AI result runs into the same measurement problem, and Adecco answered it by measuring itself. So the question to carry into a vendor conversation is a records question: what would let you compute fill rate and time to submit the same way, on the same jobs, before a rollout and after it.
Two federal actions landed on hiring paperwork.
On August 4 the Justice Department's Civil Rights Division announced a combined $3,200,000 settlement with OpenAI and its subsidiary Statsig. It resolves allegations that the companies favored workers on temporary employment visas over US workers, while recruiting through the permanent labor certification process. That process lets an employer sponsor a worker for permanent residence, but only after genuinely testing the market for US workers first.
What the department said it found was procedural throughout. The jobs never went on the external careers site, though other jobs did.
Paper applications were required for these roles while other roles were taken online, and some positions ran on the radio late at night. Fewer than ten jobs were at issue, and the department said the amount reflects the harm of shutting US workers out of well-paid tech roles.
OpenAI will pay $1,200,000 in civil penalties and set up a back-pay fund of $2,000,000. It also agreed to post those jobs publicly, accept electronic applications, train staff, revise its policies and report to the department.
The company denies the allegations, and the settlement records no admission of liability. It is the thirteenth settlement since the department re-launched its Protecting U.S. Workers Initiative in 2025.
A day later USCIS issued Policy Alert PA-2026-05. It lets officers deny a request outright where the evidence filed is missing, or does not establish that the worker qualifies. They no longer have to request more evidence first.
Officers may still ask, and they can now set shorter deadlines for a reply. The change covers filings pending or made on or after August 5, and USCIS is taking comments until September 7.
Pedagogue Systems' view. For a firm that sponsors workers, these two land in the same place. Neither turns on what anyone intended.
Both come down to what the paperwork shows: in the settlement, which channel a job ran through and how its applications were accepted, and in the policy change, whether a filing was complete the first time.
Most firms could not reconstruct the posting channel, the intake method and the advertising timing for a role they filled last quarter, and those are the three fields the department examined here.
Payrolls fell while staffing hours reached a year-to-date high.
Last week we set staffing hours against a falling hiring measure. The staffing series rose again this week, and then the July payroll release brought in a different instrument that moved the other way.
US nonfarm payrolls fell 23,000 in July, and the revisions were larger than the print itself. May came down 66,000 and June came down 37,000, leaving the two months a combined 103,000 lower than first reported. Private payrolls rose 30,000 while government fell 53,000.
Average hourly earnings were up 3.2 percent over the year, labor force participation was 61.4 percent, and temporary help services employment rose 3,400.
SIA and Bullhorn went the other way, putting US staffing hours up 10 percent year over year for the week ended July 25, a new year-to-date high. Commercial hours were up 12 percent and professional up 8 percent, with industrial up 15 percent against office and clerical down 5 percent. That indicator is a weighted sample of firms using Bullhorn, benchmarked to BLS industry composition, and its most recent weekly readings are preliminary.
Two more instruments landed the same week. ADP reported that private employers added 44,000 jobs in July, with annual pay up 4.4 percent for job-stayers and 7.0 percent for job-changers. ADP chief economist Nela Richardson said the rapid pay growth among job-changers implies supply constraints in parts of the labor market.
The ISM Services PMI read 54.1 percent, a 25th straight month of expansion, while its employment index fell to 47.4 percent and back into contraction.
Pedagogue Systems' view. These four measure different populations, so there is no particular reason to expect them to agree.
Payrolls count jobs at establishments, and staffing hours count work already delivered at firms running one software platform. ADP estimates private employment from its own payroll records covering more than 26 million employees, while ISM asks services purchasing managers about their own headcount.
A client who reads the payroll number back to you in a rate conversation is quoting one of those four populations rather than the market. The job-changer pay figure is the awkward one, because it does not fit a market that is simply cooling.
Capital moved toward agents that carry out operational work.
HappyRobot raised a $150 million Series C on August 4, led by Prysm Capital, at a post-money valuation of $1.2 billion.
Its agents handle the calls, emails and scheduling that coordinate logistics operations. The company says it serves more than 150 enterprise customers including DHL and Uber. That count came from the company, and the round announcement did not audit it.
Yellow.ai agreed on August 3 to go public through a business combination with Bluerock Acquisition Corp. Deel bought Clarity, an identity verification and deepfake detection firm, in its fifteenth acquisition.
The Yellow.ai numbers are worth separating out. The $550 million figure in circulation is pro forma equity value, at $553.5 million. The investor presentation filed with the SEC shows pro forma cash of $188.2 million, which implies a pro forma enterprise value of $365.3 million.
Pedagogue Systems' view. Equity value and enterprise value differ by net cash.
Here that gap is $188 million, on a company most readers will file away as a $550 million deal. Which figure a headline quotes matters the next time one of these turns up in a board deck as a comparable.
What we are watching.
USCIS is taking comments on its evidence-standards update until September 7. The next SIA and Bullhorn reading shows whether the 10 percent hours figure held or came down on revision, and August payrolls arrive with another round of revisions to July.
About Pedagogue Systems. Pedagogue Systems builds Cassion, a governed data foundation for staffing operations. It serves shift-based, credential-heavy operators in healthcare, industrial, and aviation staffing. Every edge is a decision. It helps staffing operators keep operational records attributable and auditable before they automate decisions.
Sources.
- The Adecco Group, second quarter 2026 results (August 6, 2026)
- Adecco Group second quarter 2026 earnings call (August 6, 2026), with Investing.com transcript coverage
- Kelly, second quarter 2026 earnings (August 6, 2026)
- US Department of Justice, settlement with OpenAI and Statsig and the settlement agreement (August 4, 2026)
- USCIS, evidence standards alert and the policy manual update (August 5, 2026)
- BLS, The Employment Situation, July 2026 (August 7, 2026)
- SIA and Bullhorn Staffing Indicator (August 4, 2026)
- ADP National Employment Report, July 2026 (August 5, 2026)
- ISM Services PMI, July 2026 (August 5, 2026)
- HappyRobot Series C announcement (August 4, 2026)
- Bluerock Acquisition Corp. investor presentation filed with the SEC (August 2026)
- Staffing Industry Analysts on Deel and Clarity (August 4, 2026)
This Brief was produced with AI assistance and adversarial review, then edited by a human before publishing. AI and people both make mistakes, so please verify anything critical independently. How we produce the Brief.