Three differently situated sources converged on a similar picture this week. The Special Competitive Studies Project task force on AI and the future of work, a senior Google executive who spent his earlier career measuring how technology reshapes economies, and a global C-suite survey from a staffing operator all pointed toward a related conclusion: AI is changing jobs faster than it is eliminating them, and the gap between adoption ambition and workforce readiness is a binding constraint on near-term gains.
The regulatory environment moved in parallel. Colorado repealed and replaced its 2024 AI Act with a narrower disclosure-and-transparency framework, effective January 1, 2027. The most-watched active American case on AI vendor liability for hiring outcomes continued to be actively litigated. Funded entrants in AI-native workforce operations took shape in healthcare. Together these signals describe a market moving from "tools in workflows" toward "agents in operations," without yet building the governance to keep that transition honest.
The SCSP Task Force releases preliminary findings.
On May 21, the Special Competitive Studies Project released the interim report of the Task Force on AI and the Future of Work. The task force is in partnership with NVIDIA, with co-chairs including U.S. Senators Mike Rounds and Mark Warner, SCSP President Ylli Bajraktari, and NVIDIA Co-Founder Chris Malachowsky. Task force members include Stanford economist Erik Brynjolfsson. The final report is due in October 2026.
Ten preliminary findings. The four most operationally consequential:
- Task-level impact precedes occupation-level impact. Jobs are bundles of tasks; AI automates or augments specific tasks before it eliminates roles. This produces a gradual reconfiguration of work without immediate changes in titles or employment levels.
- Adoption, not capability, will determine labor market outcomes. Per the task force: implementation choices, workflows, and institutional constraints shape whether AI augments or replaces. The balance is not technologically determined.
- Existing labor data cannot measure AI's effect. Employment and productivity are lagging indicators. Skill demand and hiring patterns shift before standard metrics register the change.
- Educational and credentialing systems are misaligned with rapid skill-demand changes. The front-loaded learn-then-work pattern does not fit a world where required skills change inside a career.
Pedagogue Systems' view. The first two findings name the design space Almanak is being built into. Task-level reconfiguration is the unit of work Almanak is designed to orchestrate. The adoption-shapes-outcome finding names why the data foundation matters: rules enforced at the data layer, with audit trails attached, are how an enterprise makes the implementation choices the task force describes durable. The fourth finding, on credentialing, names the reskilling gap that methodology consent and funded reskilling doctrines are designed to address.
Google's James Manyika argues the doomer timeline is wrong.
In a May 19 Platformer interview with Casey Newton, James Manyika, Senior Vice President at Google and Alphabet, offered a credentialed counter-position to recent aggressive automation predictions.
Manyika's core argument, condensed: AI has dramatically expanded the share of tasks that can be automated and extended the task duration at which automation is reliable, from roughly thirty seconds in 2017 to four-plus hours now. But occupations are bundles of coupled tasks, where the weakest link controls completion. Most jobs have couplings that resist full automation. The percentage of occupations more than ninety percent automatable has held under ten percent for nearly a decade, and most labor economists put the next-decade ceiling between two and ten percent.
Manyika is willing to bet on this. On two-year predictions made roughly two years ago that fifty percent of jobs would be wiped out: "Well, two years is up. Let's take a look. And anybody who makes that prediction for two years from now, I'm willing to take the bet."
He also offered a sharper read on the entry-level hiring contraction often blamed on AI. The sharp decline in entry-level hiring identified in widely-cited research began in October 2022. ChatGPT launched in November 2022; enterprise adoption took until 2023 to register. The macro and post-pandemic effects are doing more work in that data than AI is.
Pedagogue Systems' view. Manyika's task-level framing is independent corroboration of the SCSP preliminary finding, reached from inside Google. The "missed use" risk concept he introduces, from the UN AI governance body work, is worth surfacing: in communities where access to specialist expertise is structurally low, including rural healthcare and under-resourced regions, not using AI is itself a harm. This expands the standard risk frame from "AI may go wrong" to include "AI may not arrive." Both the arrival and the operation must be governed.
Adecco's Human Premium study quantifies the readiness gap.
Adecco Group released its Human Premium global study on May 21, 2026. The study surveyed two thousand C-suite executives across thirteen countries; respondents oversee 8.6 million workers. Forty-five percent of leaders expect AI agents in workflows within twelve months. Only thirty-six percent say their talent strategy clearly shows AI creating opportunities for employees.
The gap between leadership ambition and workforce readiness is the operative datum. A separate 2026 benchmark report from hiring automation vendor Phenom, with independent analysis from Aptitude Research, audited 219 organizations across eight industries and reported that the median company operates at roughly seventeen percent of its maximum hiring automation potential.
Pedagogue Systems' view. The seventeen percent figure is from a vendor-sponsored benchmark and should be taken as directional. What it captures is the slack between current and possible. The question for operators is whether that slack closes through unsupervised automation that creates rework downstream, or through governed automation the workforce can absorb. The latter compounds; the former does not.
Colorado repeals and replaces its 2024 AI Act.
On May 14, Colorado Governor Jared Polis signed Senate Bill 26-189, which repeals the 2024 Colorado Artificial Intelligence Act and replaces it with a narrower framework focused on disclosure and transparency. The new law takes effect January 1, 2027.
What changed. The 2024 law required a duty of care to prevent algorithmic discrimination, mandated risk-management programs, and required impact assessments for high-risk AI systems. The new law eliminates these affirmative duties along with the risk-management and impact-assessment requirements. What remains: obligations on developers and deployers of automated decision-making technology used in consequential decisions, including employment; advance notice to affected individuals; a consumer right to request meaningful human review and reconsideration after adverse decisions, to the extent commercially reasonable; three-year record retention; and enforcement exclusively by the Colorado Attorney General with no private right of action.
Pedagogue Systems' view. The 2024 framework's governance obligations (reasonable-care duty, risk-management programs, impact assessments) have narrowed substantially. What remains is a different operational shape: consumer-facing notice, the ability to respond to human-review requests on demand, and three-year retention. Systems with decision logs, notice workflows, and audit trails native to the data foundation are better positioned to operate against the remaining obligations than systems that handle them through policy documents and process diagrams.
Mobley v. Workday continues to be actively litigated.
The active litigation against Workday on the theory that its applicant-scoring software discriminates on the basis of age, with Workday liable as an agent of its client employers, continued through the spring. The case is at conditional collective certification under the Age Discrimination in Employment Act, granted May 16, 2025. On March 6, 2026, the court rejected Workday's argument that the ADEA does not protect job applicants from disparate impact. On March 30, 2026, plaintiffs filed an amended complaint reasserting California state-law disability claims previously dismissed.
A parallel case filed in January 2026 against Eightfold AI advances a different theory: that the company functions as a consumer reporting agency under the Fair Credit Reporting Act and must meet transparency obligations under that statute. One case targets outcomes; the other targets process.
Pedagogue Systems' view. Neither case has been decided on the merits. Mobley has cleared multiple procedural hurdles, including the disparate-impact ruling, and conditional collective certification is in place. The Eightfold FCRA theory has been pleaded but not yet tested. What both cases establish, taken together, is that the question of vendor liability for AI hiring outcomes is now being litigated in federal court on both disparate-impact and consumer-reporting theories. The implication for staffing operators is the same regardless of how either case resolves: the vendor relationship, the audit trail, and the credential logic must sit inside a system that can produce evidence on demand. Policy alone will not survive discovery.
AI-native entrants raise pre-seed capital in healthcare workforce operations.
Two early-stage entrants in AI-native healthcare workforce operations took shape this period. Chromie Health raised a two-million-dollar pre-seed round led by AIX Ventures to build AI agents that orchestrate hospital shift coverage through SMS-based negotiation. Worki raised two and three-quarters million dollars in pre-seed funding led by Redesign Health and Healthliant Ventures for an AI workforce infrastructure layer for healthcare, with public marketing describing built-in audit and governance and a Human Conductor oversight model.
Both are early. Pre-seed rounds at this scale demonstrate that a small set of investors will write small checks to validate a thesis; they do not establish that the category is fundable at venture scale.
Pedagogue Systems' view. In conversations over the past quarter with mid-market staffing operators, we have heard a consistent description: AI-native workforce orchestration is moving from concept toward fundable thesis, but public materials from category entrants do not yet prove the depth, portability, or auditability of the governance layer they describe. A shift-filling agent that texts qualified nurses is useful. A shift-filling agent that texts qualified nurses with credentials verified, rest requirements enforced, pay-band logic auditable, and the full transcript discoverable in a compliance proceeding is a different product. Pedagogue Systems is building Cassion as the governed data foundation for staffing operations: the substrate that makes the governance question answerable from inside the data layer rather than from inside a policy document.
What we are watching.
- The SCSP Task Force final report in October 2026, for whether the preliminary findings harden into specific policy recommendations and whether those recommendations create regulatory pressure beyond Colorado's reduced framework.
- Mobley v. Workday on merits. The case has cleared multiple procedural hurdles; the substantive question of vendor liability for AI-driven hiring outcomes is still ahead.
- Whether Q2 2026 staffing M&A produces a disclosed multiple for an AI-native staffing operations asset. Q1 volume returned, but multiples remain undisclosed. The first transparent transaction will reset valuation expectations.
- Federal-state regulatory tension on AI. Colorado's pivot is one data point. State legislatures continue to introduce AI employment legislation; federal posture has shifted toward deregulation. Operators will need to design compliance against the more demanding of whichever framework applies to a given decision.
- Workforce readiness data. Adecco's gap measurement is one current vendor-sponsored read on the constraint. Independent measurement over the next two quarters will determine whether the gap narrows through training or through unsupervised deployment that creates downstream rework.
Sources.
Special Competitive Studies Project, Task Force on AI and the Future of Work, preliminary findings, May 21, 2026. Casey Newton, interview with James Manyika in Platformer, May 19, 2026. Adecco Group, "The human premium: Leadership beyond the algorithm," May 21, 2026. Phenom and Aptitude Research, "State of Hiring Automation: 2026 Benchmark Report," based on audit of 219 organizations across eight industries. Colorado Senate Bill 26-189, signed May 14, 2026; legal analyses including Holland & Knight (May 2026) and Reed Smith (May 2026). Mobley v. Workday, Inc., Case No. 3:23-cv-00770-RFL, Northern District of California. Pre-seed funding announcements from Chromie Health and Worki.