This week, two things happened at the same time. Most people noticed one. Almost nobody connected them.
First: OpenAI released GPT-5.4. On OSWorld-Verified — a benchmark that simulates real desktop productivity tasks, not curated puzzles — it scored 75.0%. The human baseline is 72.4%. That's not a rounding error. That's AI doing general knowledge work better than the average person doing general knowledge work.
Second: Q1 2026 tech layoff data dropped. Nearly 80,000 jobs gone in the first quarter, per layoff tracking data. Of those, nearly half were explicitly attributed to AI by the companies themselves, per RationalFX data cited by Nikkei Asia and reported by Tom's Hardware. Not restructuring. Not "strategic realignment." AI, named directly, as the stated reason.
Those two facts are related. They're not a coincidence.
"Human Baseline" Used to Mean Something Different
For the last three years, every AI benchmark story came with a footnote: "performs well in controlled settings." OSWorld-Verified doesn't do controlled settings. It drops an AI into a real desktop environment — email, browsers, spreadsheets, file systems — and tells it to complete multi-step tasks. The kind of tasks that make up the actual workday of a knowledge worker.
The model scored above the human line. For the first time on a test like this.
That doesn't mean AI replaces humans across the board. It means the floor just moved. The baseline is no longer "can AI help?" It's "in which specific scenarios does a human add more value than AI does?"
That's a fundamentally different question. Most organizations aren't asking it.
The Q1 Numbers Are the Signal CEOs Were Waiting For
Here's what happened with the layoff data. CEOs have been building the case internally for AI-driven headcount reduction for 18 months. The problem: not enough public precedent to justify it to boards, investors, or the market.
That changed in Q1.
When Atlassian cuts 1,600 people, calls it explicitly AI-driven, and the stock climbed in after-hours trading — that's permission. When Reuters reported that Meta was considering cuts of up to 20% of its ~79,000-person workforce to redirect that capital into AI infrastructure, and Meta's stock climbed on the news rather than fell — the market took it as approval.
44% of U.S. hiring managers cited AI as a top anticipated driver of layoffs in 2026, per a Resume.org survey of 1,000 hiring managers.
"the short-term transition will be unusually painful compared to past technologies, since humans and labor markets are slow to react and to equilibrate."
— Dario Amodei, CEO, Anthropic, "The Adolescence of Technology" (January 2026)
The argument "AI will create more jobs than it destroys" holds over a 10-year horizon. The WEF projects 170 million new roles and 92 million displaced by 2030 — a net gain of 78 million. But the near-term is a different story. And the near-term is where Q1 2026 lives.
What This Means for How You Build Your Team
I'm a staffing company. I'm supposed to tell you to hire more people. That's not what I'm going to do here.
The calculus is this: every role you're considering adding right now should survive one question: what does this person do that GPT-5.4 can't, won't, or shouldn't do?
Not "can AI help with this role." Not "could AI do parts of this." What does a human bring that a system running at 75%+ on real-world knowledge tasks cannot replicate at the same cost?
The answers exist. They differ from what most hiring managers are using when they post job reqs.
Judgment in ambiguous situations where the wrong call costs more than the model is worth. Client relationships where trust is the product. Technical creativity that requires domain experience AI hasn't been trained on yet. Leadership of other humans when morale, context, and culture are the output.
That's a specific list. Most companies aren't building toward it.
The Agentic Shift Changes the Framing Entirely
One more piece of context: Microsoft announced its March Power Platform update this month, formally advancing an agentic architecture — systems that make decisions, learn from interactions, and execute complex processes with reduced human oversight at each step.
The 'AI as assistant' framing is retiring. 'AI as autonomous operator' is taking its place.
That means your future org chart doesn't just have fewer people in it. It has differently-shaped people. The question isn't how many. It's what kind.
The companies that figure this out in 2026 will be more competitive in 2028 than the ones that figure it out in 2028. That's how compounding works.
At VC5 Consulting, we're helping CTOs and technology leaders work through exactly this question — what does your team actually need to look like when the baseline keeps moving? If you want to think it through before your next hiring cycle, let's talk.