In the annals of technological disruption, few tales rival the quiet tragedy of the American horse. For two centuries, these noble beasts plodded on, oblivious to the hiss of steam engines chugging toward their obsolescence. Then, in a blink of equine history, they vanished.
Andy Jones, an early researcher at Anthropic - one of the vanguard AI labs challenging OpenAI's throne - recently penned a haunting reflection on his blog, drawing parallels between those forgotten herds and the accelerating specter of artificial intelligence. "Progress in engines was steady," Jones writes.
"Equivalence to horses was sudden." As AI investments balloon and models like Claude eclipse human expertise overnight, Jones warns: We're not the inventors this time - we're the horses.
Jones' post, a five-minute lightning talk from summer 2025, isn't mere nostalgia. It's a dispatch from the front lines of AI's workforce revolution, where steady capital pours yield abrupt existential shocks. With global AI data center spending equivalent to 2% of U.S. GDP - roughly $560 billion in 2025, up from $342 billion in hyperscaler capex alone - the industry's trajectory mirrors the engine's inexorable grind.
Yet, as Jones attests from his perch at Anthropic, the human toll arrives not in decades, but months. In an era where AI drove 92% of U.S. GDP growth in the first half of 2025 through data center booms, the question looms: How long before the machines outpace not just our tools, but us?
The Steam Engine's Silent Siege: 200 Years to Irrelevance
Picture 1920 America: 25 million horses thundered across farms and streets, powering plows, pulling wagons, and embodying rural might. These equines, numbering one for every 4.6 people, were the backbone of an agrarian economy - indispensable, irreplaceable.
Steam engines, patented in rudimentary form by Thomas Savery in 1698 and refined by James Watt in 1769, had been chugging along for over two centuries.
By the 19th century, they drove factories and locomotives, improving efficiency by about 20% every decade through tweaks in pressure, materials, and design. Yet horses, grazing contentedly, noticed nothing. From 1850 to 1910, U.S. equine populations swelled from 4.8 million to a peak of 26.5 million, fueled by westward expansion and the Civil War's demand for cavalry mounts.
The tipping point? The interwar mechanization boom. Ford's assembly line slashed tractor costs, dropping from $1,000 in 1915 to under $300 by 1930. By 1930, gasoline engines outperformed horses in cost and output, but the real cull came post-Depression: Between 1930 and 1950, 90% of those 25 million horses vanished, plummeting to just 2.5 million by mid-century.
Tractors, trucks, and combines didn't just compete - they eradicated. By 1951, only a million horses remained, relegated to racetracks and ranches. World War I's export of 500,000 draft animals to Europe accelerated the slide, but it was the tractor's ubiquity - rising from 600,000 in 1920 to 2.5 million by 1940 - that sealed the fate. Jones nails it: Progress plodded steadily; displacement struck like lightning.
Checkmate: From Underdogs to Overlords in a Decade
Fast-forward to the chessboard, where silicon minds mirrored this pattern with eerie precision. In 1985, as IBM's Deep Thought notched its first tournament wins, computer Elo ratings hovered around 2000—solid club player level, but fodder for grandmasters.
Over the next 40 years, engines like Stockfish and AlphaZero climbed relentlessly, gaining about 50 Elo points annually through brute-force search trees, neural nets, and hardware leaps.
By 2000, the top program rated ~2600 - enough for a human grandmaster (average ~2600 Elo) to claim victory in 90% of matches, as Garry Kasparov trounced Deep Blue 3-2 in their 1996 rematch.
Enter the aughts: Moore's Law turbocharged parallel processing, and by 2006, Hydra edged past 2800 Elo, tying top humans. The inflection? 2010, when Rybka and friends surged to 3200+, flipping the script: That same grandmaster now lost 90% of games, a reversal as stark as a pawn's promotion to queen.
No scandals or scandals - just exponential compute. From 1985's 2000 Elo to 2022's 3600+, the curve was a gentle sigmoid, but human parity hit like a fork: sudden, total. As Jones observes, "Progress in chess was steady. Equivalence to humans was sudden."
Claude's Quiet Coup: Six Months to Surpass a Specialist
Now, the mirror held to AI itself. Global datacenter outlays for machine learning - servers, GPUs, cooling - hit $50 billion in 2024, ballooning to $80-100 billion in 2025, equivalent to 2% of U.S. GDP and doubling yearly on megadeals like Microsoft's $80 billion hyperscaler pledge. This steady infusion powers models like Anthropic's Claude, which in 2024 began nibbling at the edges of human cognition.
Jones, hired in Anthropic's infancy, spent much of 2024 fielding a deluge: 4,000 technical queries monthly from fresh recruits - debugging code, unpacking architectures, troubleshooting edge cases.
Teamed with a handful of veterans, it was rote but vital scaffolding for a startup scaling to thousands. December 2024: Claude 3.5 Sonnet steps in, handling "some" questions.
Six months on - by mid-2025 - 80% of Jones' inbox evaporates. Claude now fields 30,000 queries monthly, eightfold the human throughput, at a whisper of the cost: pennies per inference versus a senior researcher's $200,000 salary.
This wasn't gradual erosion; it was a flash flood. Query volume didn't taper - it imploded as Claude's accuracy spiked from 60% to 95% on domain-specific probes, per internal benchmarks. Horses endured two decades of parity before the axe fell; chess masters, a mere ten years. Jones? Six months. "It took me all of six months to be surpassed," he laments, "by a system that costs one thousand times less."
Galloping Toward the Abyss: AI's Reckoning
Jones closes with a plea laced in fatalism: "I very much hope we'll get the two decades that horses did. But looking at how fast Claude is automating my job, I think we're getting a lot less." His post, dated December 8, 2025, arrives amid Anthropic's $18.4 billion valuation and Claude's enterprise pivot, where AI now drafts patents and simulates experiments, slashing R&D cycles from weeks to hours.
The irony? We're pouring trillions into this singularity-lite, with U.S. AI infrastructure alone projected at $3-4 trillion by decade's end. Yet as engines felled horses and algorithms felled kings, Claude doesn't just assist - it supplants, quietly rewriting job descriptions in labs worldwide.
In 1920's amber glow, 25 million horses grazed, deaf to the factories' hum. Today, in server farms humming at 2% of GDP, we code on - hoping our two centuries buy more than six months. But history whispers: Steady progress devours suddenly. The herd thins fast.
Also read:
- From Fortress of IP to AI Playground: Disney's $1 Billion Leap into OpenAI's Sora Universe
- OpenAI's GPT-5.2 Revolution: Empowering Professionals with Frontier AI Intelligence
- Substack's Native Sponsorships: Revolutionizing Creator-Brand Partnerships in the Newsletter Era
Author: Slava Vasipenok
Founder and CEO of QUASA (quasa.io) - Daily insights on Web3, AI, Crypto, and Freelance. Stay updated on finance, technology trends, and creator tools - with sources and real value.
Innovative entrepreneur with over 20 years of experience in IT, fintech, and blockchain. Specializes in decentralized solutions for freelancing, helping to overcome the barriers of traditional finance, especially in developing regions.
