Three things happened in the last full week of May that usually move on separate tracks: a large talent platform put AI at the center of a record profit, a staffing-software incumbent shipped agentic "digital workers" into the recruiter's daily workflow, and two states moved on AI-employment rules, one of them repealing and replacing a law it had not yet enforced. Read together, they describe a single direction. AI is moving from assisting the work to performing parts of it, and the rules governing that shift are a moving, state-by-state patchwork rather than a settled framework. For staffing operators, the useful question is no longer whether to adopt AI. It is whether the AI operates inside policy the operator can enforce and show.
A talent platform credits AI for record results, at the marketplace layer.
According to Recruit Holdings' fiscal-year results, the parent of Indeed and Glassdoor reported record numbers: operating profit up 28.5 percent, profit attributable to owners up 21.6 percent, and return on equity climbing from 22.6 percent to 31.0 percent, with guidance for 9 percent revenue growth in the year ahead. Management credited AI, pointing to matching models and improved monetization on Indeed, where US revenue per job posting rose about 17 percent even as hiring demand stayed soft.
Pedagogue Systems' view. Management credited AI, and the result does not isolate AI's contribution from pricing, marketplace mix, and cost discipline. The gains are concentrated in a job marketplace, and the same segment announced a workforce reduction of about 6 percent, roughly 1,300 roles, in 2025. That is AI alongside monetization and efficiency at the matching layer, which is a different thing from AI producing margin inside the delivery of credentialed, shift-based labor. The result is a signal of what AI can do where the product is matching. It does not yet answer what AI can do where the product is a qualified person showing up to a regulated shift. That second question is the one most operators need answered.
Agentic capability is shipping into the daily workflow.
Bullhorn used its Engage Boston conference on May 28 to make its Amplify "digital workers" generally available to Amplify customers, adding skills it calls Prospect, Verify, Audit, and Transcribe along with a conversational command layer, and describing the shift as moving from a system of record to a system of action. Separately, Workday introduced Adaptive Decision Intelligence, a capability in Workday Adaptive Planning that lets finance teams ask questions in natural language, model scenarios, and commit approved decisions into the plan, which Workday says runs inside the platform's existing permissions and keeps an audit trail of the data and assumptions behind each scenario.
Pedagogue Systems' view. These are two different motions, and both are strong at what they do. Bullhorn is embedding agentic automation in execution workflows, where recruiters and salespeople work. Workday is embedding AI in enterprise planning, where finance teams work. The common thread is that the frontier is moving from AI that suggests to AI that acts, and once software can initiate or commit work the governing questions are the same in both places: what is this agent permitted to do, who approved that, and can you reconstruct it later. In credential-heavy, regulated work, a system of action is most useful when the policy it acts within is enforced rather than assumed, and when the record of what it did is a property of the system rather than a report someone runs afterward.
AI depth, not AI presence, tracks growth.
StaffingHub's 2026 State of Staffing benchmarking report, drawn from a survey of agency leaders, found that the share of agencies reporting revenue declines fell as AI use deepened: about 41 percent among agencies using AI in one or two processes, 34 percent at three or four, and 31 percent at five or more, with heavy adopters more than twice as likely to land in the high-growth tier. The strongest correlations were practical, embedded use cases: job-description generation and reporting and analytics, each about 2.7 times more common in growth agencies, recruiting chatbots at 2.3 times, and candidate qualification at 2.2 times. The report ties growth more to operational maturity, measured as a scorecard of operating habits like weekly KPI review and documented procedures, than to size or speed metrics. About 46 percent of surveyed agencies still use no AI at all.
Pedagogue Systems' view. This is about workflow depth, and it is correlational, not causal. Deeper AI use travels with operating maturity, and maturity may be doing much of the work. The honest implication for governance is not that governance produces the growth. It is that depth and obligation rise together. The more workflows an operator hands to AI, the larger the surface that has to be governed, disclosed, and reconstructable. The agencies pulling ahead are the ones redesigning process around the technology, and redesigned process is what needs an enforcement layer underneath it.
The regulatory map keeps getting rewritten, and no federal statute sits beneath it.
In May, Connecticut enacted Senate Bill 5, the Artificial Intelligence Responsibility and Transparency Act, a broad online-safety and AI law whose employment provisions take a disclosure-focused approach: notice requirements for automated employment-decision tools, a clarification that using AI is not a defense to a discrimination claim, and an AI-disclosure obligation tied to layoff notices, phasing in from October 2026. Colorado signed Senate Bill 26-189, which repeals and replaces its 2024 AI Act with a narrower notice-and-transparency framework, with covered obligations taking effect January 1, 2027. Enforcement in Colorado is currently paused: a federal court stayed enforcement of the prior law in late April, the stay extends to the replacement, and the attorney general has said he will not enforce either until his rulemaking is complete. These join New York City's bias-audit and notice rule and Illinois's AI-disclosure requirement. No enacted federal AI-employment statute sits beneath them.
Pedagogue Systems' view. The laws differ in mechanism, not just in detail. New York City runs on bias audits and candidate notice, Illinois on disclosure and discrimination liability, Connecticut on disclosure inside a broad online-safety law, and Colorado on notice plus a retained record plus a right to meaningful human review of adverse decisions. They do not share one rulebook. What they share is direction: tell people when an automated tool is used, disclose its role when an outcome is adverse, and keep records long enough to show what happened. The last two years argue against planning around a single, near-term uniform framework. The durable response is an architecture that produces notice, disclosure, and a retained, reconstructable record by default, so that complying with the next regime is a configuration question rather than a rebuild.
The recovery is shallow and selective, and it rewards specialization.
The ASA Staffing Index sits modestly above year-ago levels, up roughly 4.6 to 4.8 percent, while still below pre-2024 levels and dipping slightly in the latest week. Bullhorn's near-real-time platform data shows staffing hours at 2026 highs across most segments, led by light industrial at about 9 percent over last year and commercial at about 6 percent, with office and clerical lagging, and temporary IT openings up about 22 percent year over year. SIA's forecast puts the US healthcare staffing market near 39.4 billion dollars in 2025, down about 6 percent, climbing out of a deeper trough with a modest recovery expected this year. On the deal side, merger roundups put the start of 2026 at its strongest pace since 2022, clustered in IT, executive search, and other specialized segments. Reported valuations remain wide: generalist staffing is often cited in a directional range of roughly half to one times revenue, but multiples vary materially by margin profile, growth, and contract mix, and specialized businesses generally command higher multiples than generalists.
Pedagogue Systems' view. This is not a volume recovery that rewards generalist scale. It is a selective one that rewards specialization, operating discipline, and cost-to-serve, concentrated in credential-heavy and contingent work. That is the terrain where Pedagogue Systems expects governed AI to earn its keep, and it is a thesis the company is testing rather than a market result it can point to yet. The mechanism is specific: in regulated, credential-heavy work, governed AI pays off if it reduces exception handling, repeated credential checks, and the time it takes to assemble an audit, which is where the cost of an unauditable automated decision is highest and the premium for doing it cleanly is most visible.
What we are watching.
- How Colorado's attorney-general rulemaking resolves and when the stayed enforcement lifts, particularly how meaningful human review and material influence get defined in practice.
- Connecticut's automated employment-decision obligations as their October 2026 effective dates approach.
- The US Department of Labor's proposed joint-employer rule, whose comment period closes June 22, 2026, and what it implies for how client and agency relationships are structured.
- The first wave of independently reported results from agentic deployments now reaching production, as distinct from vendor-reported figures.
About Pedagogue Systems.
Pedagogue Systems builds Cassion, a governed data foundation for the staffing industry. Cassion enforces state machines, access controls, a credential engine, and an audit substrate at the database layer, so that AI can take on more of the work while every consequential action stays permitted, attributable, and reconstructable. The aim is straightforward: make AI safe to use in regulated, credential-heavy work by governing it where the data lives.
Sources.
Recruit Holdings fiscal-year results and workforce-reduction disclosure (Recruit Holdings investor materials and newsroom; HR Dive). Bullhorn Amplify announcement (Bullhorn, Engage Boston, May 28). Workday Adaptive Decision Intelligence announcement (Workday, May 27). 2026 State of Staffing Benchmarking Report (StaffingHub, with summary analysis by Avionte). Connecticut Senate Bill 5 analyses (Littler, ArentFox Schiff, Shipman and Goodwin, CBIA). Colorado Senate Bill 26-189 analyses (Buchalter, McDermott, Snell and Wilmer, Crowell and Moring). ASA Staffing Index (ASA). SIA US healthcare staffing forecast. Q1 2026 staffing merger roundups. US Department of Labor joint-employer proposed rule.