The Quiet Layoff

Here's a thought experiment: if an AI agent can file your expense reports, update your CRM, and draft your Jira tickets, how many Salesforce seats does your company actually need? Goldman Sachs just put that question in writing, issuing a research note warning that "agentic AI" poses a structural threat to the entire SaaS business model. Hedge funds are already dumping software stocks.

The logic is almost insultingly simple. Enterprise software companies charge per seat. AI agents don't need seats. When one agent can do the work of three human users across ServiceNow, Salesforce, and Slack, the math doesn't favor the incumbents. Goldman's note suggests this isn't a 2030 problem—it's already showing up in pipeline forecasts.

What makes this significant isn't the prediction itself (plenty of people have said this), but who's saying it. When the same bank that just deployed AI to 12,000 of its own people tells investors to rethink software stocks, that's not speculation. That's a field report.

The difference between "pilot" and "production" is the difference between curiosity and commitment. Goldman Sachs just crossed that line, rolling out Anthropic's Claude to over 12,000 developers and thousands of back-office staff. Initial pilots showed a 20% boost in developer productivity. That number was apparently convincing enough to bet the entire engineering org on it.

The use cases tell you where the jobs are changing: trade accounting, compliance checks, onboarding workflows. These aren't moonshot experiments. They're the repetitive, judgment-light tasks that employ hundreds of thousands of people at every major financial institution. When a regulated, risk-averse bank like Goldman goes all-in, the compliance objection—the last defense of "we can't use AI for that"—is effectively dead.

The harder question is what happens to the productivity surplus. A 20% boost in developer output can mean 20% more features—or 17% fewer developers. History suggests companies eventually choose door number two.

McKinsey has never met a transformation it couldn't quantify. This week's contribution: generative AI will unlock between $50 billion and $70 billion in value for the insurance industry. The primary impacted areas? Marketing, sales, customer operations, and software engineering. If you work in insurance and your job involves writing, reviewing, or processing anything, this report is about you.

The report's real significance is in the language shift. McKinsey isn't advising firms to "explore" or "experiment." The recommendation is to move from pilot to scale. That's consulting-speak for "the decision has been made, you're just behind." When the industry's most influential advisory firm tells every insurer in the world to accelerate AI deployment, the job displacement timeline compresses from "eventually" to "actively."

Worth noting: $50–70 billion in "value" doesn't mean $50–70 billion in new revenue. Most of it comes from doing the same work with fewer people. The polite term is "efficiency gains." The blunt term is payroll reduction.

Amazon is eliminating approximately 16,000 corporate and tech-facing jobs globally. This is the second major round in three months, bringing recent totals to around 30,000. CEO Andy Jassy's internal memo was unusually direct: "We are flattening our structure to move faster and doubling down on our biggest growth areas, specifically generative AI."

What's notable here isn't the size—Amazon cuts people regularly. It's the framing. Previous rounds were described as "right-sizing" or "post-pandemic corrections." This one explicitly names AI as the reason. The restructuring is "aimed at flattening management and reallocating resources toward artificial intelligence priorities." They're not hiding it anymore. They don't need to.

The template is now clear: flatten the middle, automate the routine, and redirect the savings to AI infrastructure. Every Fortune 500 CEO read that memo. The ones who haven't started their own version are writing theirs now.

If you want to know whether companies are actually hiring, don't ask the companies. Ask Workday. The HR software giant processes hiring for a significant chunk of the Fortune 500, and their latest report is bleak: 66% of CEOs plan to cut or freeze staff levels in 2026 due to AI efficiency gains. Workday itself described this as an "existential threat" to its own growth model.

Think about what that number means. Two-thirds of chief executives are planning to maintain or shrink their workforce, not because of recession, not because of a demand shock, but because they believe AI can absorb the work. This is a structural shift in hiring philosophy, not a cyclical dip. The job market isn't weak—it's being deliberately compressed.

The irony runs deep. Workday's entire business exists because companies need software to manage large, complex workforces. If those workforces get permanently smaller, Workday's addressable market shrinks alongside them. The canary is choking on the same gas it's supposed to detect.

While the private sector races to replace workers with AI, one state has decided to regulate how AI replaces the replacements. Colorado's new law governing AI in hiring and employment decisions took effect on February 1, making it the first state to enforce strict rules on algorithmic decision-making in HR. Employers must now notify candidates when AI is used in screening, and conduct regular bias audits of their systems.

The law covers "high-risk" AI systems used for hiring, promotion, and performance evaluation. That's a broad net. If your company uses HireVue for video screening, Workday's AI for resume filtering, or any automated tool to rank candidates, you're now subject to disclosure and audit requirements in Colorado.

Here's the strategic implication: because the federal government hasn't acted, Colorado is setting the de facto national standard. Any employer hiring remotely (which is most of them now) needs to comply if a single applicant is in Colorado. The EU's AI Act took a similar path—one jurisdiction's rules became everyone's floor. The companies building AI hiring tools just got their first meaningful constraint. Whether it's enough to matter is another question entirely.