Two developments this week force a question that labor economists have mostly avoided: what happens when AI stops assisting human workers and starts replacing the judgment calls at the core of their jobs?

The first development is a legal one; the second, a technology one.

In the first instance, plaintiffs in the landmark case Mobley v. Workday, Inc. filed an amended complaint in late March 2026. The complaint revived California state claims and a physical disability discrimination count after a federal judge’s ruling kept core federal age discrimination claims alive.

The case involves Derek Mobley, an African-American man over 40 with anxiety and depression. He applied for more than 100 jobs through companies using Workday’s AI-powered applicant screening tools. He was rejected every time, sometimes within minutes of applying.

A federal court has allowed discrimination claims against Workday’s AI screening tools to proceed on an unprecedented “agent” liability theory. The case is advancing primarily on age discrimination under the Age Discrimination in Employment Act. The outcome of the case could impact the lives of hundreds of millions of job applicants.

The case is the first major test of whether AI vendors — not just employers — can be held directly liable for algorithmic hiring bias under federal anti-discrimination law.

The legal significance here is structural, not just factual. AI hiring tools are used by thousands of companies as a cost-saving measure.

87% of companies are now using AI-driven tools to save time and money in the hiring process. When those tools discriminate, they do so at scale, across every employer using the same platform simultaneously.

The Workday case tests a theory that has major consequences for anyone in the job market over age 40: that the vendor of the discriminating tool shares legal liability with the employer.

If the court upholds that theory, it opens class-action exposure for every major HR software company operating AI screening at scale.

A University of Washington study found that recruiters using biased AI tools mirrored the tool’s inequitable choices up to 90% of the time. The study illustrates that both the AI vendor and the employer carry liability.

If your company runs AI applicant tracking systems without documented bias testing, adverse impact analysis, and a human-in-the-loop override, the Board is sitting on a legal exposure it did not budget for.

The second development connects to the first at the level of capability.

OpenAI’s GPT-5.4, released in early March, achieves a new state of the art on GDPval. GDPval is a benchmark that tests AI agents on well-specified knowledge work across 44 occupations.

GDPval measures real work products: sales presentations, accounting spreadsheets, urgent care schedules, manufacturing diagrams. It is not a test of abstract reasoning. It is a test of the actual output that knowledge workers produce on a given Tuesday.

OpenAI’s 5.4 model matches or exceeds industry professionals in 83% of comparisons. That’s up from 70.9% for its predecessor, GPT-5.2.

On an internal benchmark simulating tasks that a junior investment banking analyst might perform, GPT-5.4 scored 87.3%, up from 68.4% for GPT-5.2. Tasks include spreadsheet modeling, scenario analysis, and data extraction.

The improvement represents a 28-point gain in under four months. The rapid improvement suggests accuracy could reach human professional-levels in the high 80s or low 90s within the next year if the trend holds.

So, AI tools are now being used to screen job applicants, manage HR workflows, and conduct tasks previously performed by junior knowledge workers. The Workday case reveals that those tools carry discrimination risks that no one has adequately audited.

The GPT-5.4 benchmark results reveal that those tools are improving fast enough that the category of “tasks requiring human judgment” is shrinking.

AI is not merely replacing tasks, but deconstructing occupations into automated and human-centric components.

“Deconstructing occupations” describes something more precise than replacement. A job that used to require a human is being split into its component tasks. Some tasks go to AI.

The remaining tasks stay with humans; those activities that require judgment, relationship management, or physical presence. The human does fewer things. Their employer employs fewer people.

In the Gulf region, financial institutions are deploying autonomous agents to handle 80% of routine compliance and reporting. The agents enable human staff to focus on high-stakes advisory and regional strategy.

The most notable change in 2026 is the move toward agentic workflows. Modern AI agents can now coordinate across departments. They can manage complex projects with minimal human oversight.

Future Forwarded is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.

According to PwC, roughly 44% of core worker skills have been disrupted in the last 24 months. The new gold standard for the global workforce includes:

* strategic prompting and orchestration

* the ability to manage multiple AI agents simultaneously

* AI ethics and bias mitigation

* and complex emotional intelligence , or EQ

So, the Workday case is working through a legal system designed for a world where a human manager made hiring decisions. The AI tools now making those decisions at scale were designed by engineers optimizing for efficiency.

The discrimination they produce is not intentional. That makes it harder to detect, harder to litigate, and harder to remedy.

Colorado’s AI law, set to take effect at the end of June 2026, will require employers in the state that deploy high-risk AI systems to take “reasonable care” to protect consumers from discrimination.

HR software that screens candidates, scores performance, or ranks employees is classified as high-risk AI under Colorado’s law. Colorado is the first state to impose that standard.

Whether other states follow suit before the tools spread further is an open question. The outcome has real consequences for every worker who will be filtered by an algorithm in the next job search.

A compilation of the Substack articles examining how the invasion already happened. You just weren’t invited. $9.95 flat fee for the bundle (PDF, ePUB), no subscription required. 2-hr reading time.

Get full access to Future Forwarded at futureforwarded.substack.com/subscribe

Get full access to The AI/Labor Report at ailabor.substack.com/subscribe