Last Updated: June 4, 2026

The 36-Month Miracle

⚠ June 2026 Update — The Miracle Went Autonomous

When this page was written, the equation read "intelligence became cheap." It's now also become autonomous and recursive. In April 2026, Anthropic launched Managed Agents. Around the same time, an engineer at Rakuten had Claude Code implement a feature across a 12.5-million-line codebase in a single 7-hour autonomous run at 99.9% accuracy. METR's autonomy benchmark measured Claude Opus 4.6 at a ~14.5-hour 50% task horizon in Feb 2026 — and that horizon has been doubling every 4–7 months.

Jack Clark (Anthropic policy lead) put a 60%+ probability on a frontier AI fully training its own successor by end of 2028 in his Oxford Cosmos Lecture, May 2026 — the textbook definition of recursive self-improvement. Anthropic's own internal timeline (per Redwood Research's analysis of their OSTP filing) projects >10x research engineer acceleration by end of 2026 and possible full AI R&D automation by early 2027.

The multiplication factors below were already staggering. They are now the conservative case. Companion piece: AI Is Getting Really Serious.

Intelligence used to be the most expensive AND rarest commodity on the planet.
In 3 years, we turned it into a utility cheaper than tap water.

In early 2023, AI wrote at a sixth-grade level. By 2025, it was producing Post-Doc research. Then came the Agents—autonomous digital employees that don't just talk, they act.

Capability

60%

smart high-schooler

Speed

heavy human review

Scale

parallel agents

Early AI (2023)

Lower capability means more human review and AI re-thinking passes before output is usable—eating into the speed advantage. At near-human expertise, review becomes negligible. Real-world overhead also depends on task importance, model cost, and use case—we're simplifying here to show the trend.

The Chronology of Displacement

First

The "Inbox Man"

Data entry, routing, routine compliance. Replaced instantly—zero friction, no physical presence needed. Just inputs, rules, outputs. AI does it better and never asks for vacation.

Slower

The Physical Frontier

Code moves at light speed. Moving boxes requires a body. Trades and physical work persist—not because they're simple, but because they need presence.

BUT...

Last

The "Human Shield"

CEOs aren't paid for logic—they're paid for status. The person the tribe trusts, the law can sue, the face that represents "us." They become Wise Idea Coordinators.

The "Abundance" Tensions

AI software gets cheaper every quarter as model sizes shrink, costs per token fall, and open-source models proliferate — while AI hardware decelerates against datacenter buildout delays, denied building permits, copper and transformer shortages, and three-year backorders

The Economic Feedback Loop

Spreadsheet Logic "Lay off 50% of staff to maximize quarterly margins."

Social Reality 50% unemployment = no customers. The company dies with the community.

The Abundance Question

The Vision AI creates so much wealth that everyone benefits through redistribution.

Structural Reality Cash doesn't help if housing, energy, healthcare remain scarce. Abundance = access.

The Species-Wide Re-Tasking

"For decades, we invented 'Bullshit Jobs' to fill 40 hours a week because we didn't know how to organize anything else."

AI removes the Bullshit. It doesn't leave us idle; it re-tasks us.

Cleaning the oceans and land

Retrofitting homes for energy independence

Rebuilding infrastructure

High-touch elder care

Community mental health

Regenerative agriculture

AI Will Make Strong Communities — By Necessity

This isn't a government program. This is how people survive. Hover to see how.

🏠

Housing Crisis

Hover to see solution

AI coordinates 500 neighbors to pool resources, navigate zoning, and structure community land trusts—paperwork that took years, done in days. Intelligence does the paperwork

🌉

Broken Infrastructure

Hover to see solution

Agent swarms analyze requirements, source materials, coordinate volunteers, handle permits. Communities become self-repairing organisms. Bypass bureaucracy

🏥

Healthcare Gaps

Hover to see solution

Post-Doc level AI provides monitoring, handles insurance bureaucracy, coordinates care. Expert knowledge democratized. Expertise goes ambient

📚

Education Inequality

Hover to see solution

Every child gets an infinitely patient tutor, available 24/7, adapting to their learning style. Education stops being scarce. The great equalizer

About the author

I'm Scott Covert. I'm 60, based in Peterborough, Ontario, and I've been using frontier AI for real work — research, writing, business plumbing, medical advocacy, financial analysis — since early 2023. I write about this from the inside, not as an observer. I built AI Over 50 for adults my age who'd benefit from these tools more than anyone but get the worst instruction.

The numbers on this page aren't projections. They are public benchmark results (METR, SWE-bench, ARC-AGI) and direct quotes from frontier-lab announcements. Every claim is sourced below. If you find one that doesn't hold up, email me: [email protected] — I update when I'm wrong.

More work: Substack · Medium · AI Over 50

Sources cited on this page

Anthropic Managed Agents (April 2026) — the-decoder.com — cloud-hosted autonomous agents; plus Claude Code's Rakuten 12.5M-line, single 7-hour run.
METR Autonomy Time Horizons — metr.org/time-horizons — Claude Opus 4.6 measured at ~14.5-hour 50% task horizon (Feb 2026); horizon doubling every 4–7 months.
Jack Clark, Anthropic — Oxford Cosmos Lecture (May 20, 2026) — Axios coverage — "60%+ chance a frontier AI fully trains its successor by end of 2028."
Anthropic OSTP filing — Redwood Research analysis — blog.redwoodresearch.org — mid-2026 ~3x research engineer acceleration; end of 2026 >10x; possible full AI R&D automation early 2027.
SWE-bench Verified leaderboards (June 2026) — swebench.com — Claude Mythos Preview 93.9%, Claude Opus 4.8 88.6%, Claude Opus 4.7 87.6%.
Claude Opus 4.7 announcement (April 16, 2026) — anthropic.com/news/claude-opus-4-7 — SWE-bench Pro 64.3%, GPQA Diamond 94.2%.
OpenAI GPT-5.5 ("Spud") launch (April 23, 2026) — openai.com — Terminal-Bench 2.0 82.7%, ARC-AGI-2 85.0%, FrontierMath T1-3 51.7%.

Methodology note: the rotating equation above (capability × speed × scale) is a conceptual frame to communicate compounding effect, not a measured benchmark. The sourced data points above ground the qualitative claims in the article.

The spreadsheet is being rewritten. You aren't a victim of the math; you're the one who decides what the variables are for.