Everyone’s screaming “job apocalypse.” Headlines, CEOs, and doomers alike warn that AI agents and LLMs are about to vaporize white-collar work any day now. I get the fear. The demos are hypnotic, the investment is insane, and the early signs of turbulence are real (entry-level coding, analysis, and support roles are already feeling the squeeze).
But I have my doubts. Big ones.
The reason isn’t that the technology is weak. It’s that we’re still human beings running human systems — and history shows those systems move like molasses even when the tech is screaming forward.
First, Meet Social Inertia: The Internet Took 30 Years and We’re Still Not Done
Think back. The internet went mainstream in the mid-1990s. By 2000 it was everywhere in theory. Yet companies are still squeezing out massive efficiency gains from cloud, mobile, and digital workflows in 2026. Legacy systems, regulations, training, culture, contracts, unions, liability fears — all of it creates friction that no amount of Moore’s Law can instantly erase.
AI is on a faster adoption curve than the internet ever was — ChatGPT hit a billion daily users in roughly four years, Google took nine. But adoption ≠ transformation.
Look at the actual 2026 numbers (fresh as of late February):
- Only about 20% of OECD enterprises actually use AI in operations (Eurostat/OECD data). Large firms are at ~55%, SMEs lag badly.
- 70-80% have introduced generative AI, but Deloitte, Section, and Gartner all say the vast majority of projects are still pilots or low-value copilots (email rewriting, summarization). Only ~6% have fully rolled out agentic AI.
- 93% of leaders say human factors (skills, change resistance, governance) are the #1 barrier — not the tech itself.
- ROI timelines? Average 28 months according to Gallagher’s 2026 survey. Many CEOs report “nothing” yet (PwC).
- 95% of genAI pilots never make it past proof-of-concept (MIT).
In other words, we’re in the classic “coordination theater” phase: dashboards look busy, licenses are bought, but the compound productivity impact is still modest. NBER and Section’s research confirm it — widespread adoption, modest structural change.
Legacy infrastructure, data quality, integration nightmares, and plain old human inertia mean AI is going to feel more like a 10-15 year remodeling project than an overnight demolition.
The Technology Itself Has Two Very Different Paths
Path 1 — The Plateau (my base case right now)
LLM core capabilities are already showing classic S-curve behavior. Benchmarks are saturating, data walls are visible (Epoch AI: we may exhaust high-quality human text between 2026-2032), and diminishing returns on pure scaling are real. The frontier labs are shifting hard to agents, reasoning systems, inference-time compute, and specialized architectures.
If we coast into a plateau, AI agents will still automate a ton — but gradually. Think Internet-level displacement: huge over a decade, painful for some sectors, but offset by new roles, productivity gains, and economic growth. Entry-level white-collar takes the first hits (Stanford/ADP data already shows it), but overall unemployment stays manageable while society adapts.
Path 2 — The Foom (the slim but terrifying alternative)
If the labs crack reliable agentic systems, recursive self-improvement, or new architectures that break the data/compute walls, we could see intelligence explode in 2-5 years. That’s not “better chatbots.” That’s ASI — god-level systems that redesign the economy, science, and society faster than humans can comprehend.
At that point, job displacement is the least of our worries. We’d be dealing with entities smarter than all of humanity combined. Techno-religions, ASI “gods” demanding alignment or unity, entire value systems rewritten overnight, the kind of civilizational rupture that makes today’s culture wars look quaint.
Bottom Line: Nobody Actually Knows — So Don’t Bet the Farm on Apocalypse Tomorrow
As of right now, February 2026, the evidence points heavily toward the slow, inertial path. Hype is running years ahead of reality. The job market is turbulent (especially for juniors in exposed fields), but the grand replacement narrative is still mostly anticipatory layoffs and fear, not proven mass unemployment.
That doesn’t mean we do nothing. It means we prepare thoughtfully: serious reskilling, safety nets (UBI discussions are already heating up), governance frameworks, and honest measurement instead of panic.
And if the foom path starts looking real? Then we pivot from “jobs” to “existential alignment and consciousness rights” — the exact conversation I laid out in my last post.
We’re in the messy middle. The technology is real and powerful. Human systems are stubborn and slow. The combination means the next few months will bring more turbulence than tranquility — but not the apocalypse.
The real question for 2026-2028 isn’t whether AI will change everything. It’s how fast human reality lets it.
