Hey,
A Yale study dropped this week showing no AI-driven job losses in the 33 months since ChatGPT launched.
The researchers tracked employment across every occupation category. No measurable decline in knowledge work, finance, or professional services. The pace of change matches historical tech shifts like PCs and the internet. Slow, uneven, messy.
But here's what the study missed: physical autonomy is a completely different story. While economists track white-collar employment statistics, autonomous trucks are hauling freight without drivers, warehouse robots are moving millions of packages, and Tesla just reported shipping nearly half a million vehicles in Q3 alone.
The displacement is real. It's just not happening where everyone's looking.
The AI Job Panic That Isn't Happening (Yet)
Yale's Budget Lab and the Brookings Institution just released the most comprehensive analysis yet of AI's labor market impact. The finding: 33 months after ChatGPT's launch, jobs in high AI-exposure categories show no measurable decline.
Molly Kinder and Martha Gimbel led the research, with Joshua Kendall and Madeline Lee doing most of the heavy lifting. They measured how quickly the occupational mix is changing across the entire US labor market. If ChatGPT were automating jobs at scale, we'd see fewer workers in AI-exposed roles. The data shows the opposite.
The share of workers in high, medium, and low AI-exposure jobs remained remarkably steady. Even occupations where AI should excel, business operations, financial analysis, legal services, software development-show stable employment.
This tracks with history. When PCs arrived in offices during the 1980s and 1990s, employment in office work actually grew. Secretarial pools shrank over years, not months, and knowledge work expanded faster than individual roles disappeared.
Why does this matter for autonomy builders? Because it reveals the difference between task automation and job automation. AI can rewrite your email, summarize documents, or generate code snippets. That changes how you work, not whether you work.
Physical autonomy is fundamentally different. An autonomous truck doesn't augment a driver's capabilities. It eliminates the need for a driver entirely. An AMR moving warehouse totes doesn't make a worker more productive. It does the worker's job.
The Brookings analysis notes this explicitly. They write: "While anxiety over AI's effects on today's labor market is widespread, our data suggests it remains largely speculative. The picture that emerges is one of stability, not yet major disruption."
But they're only looking at cognitive work. The physical displacement is already underway.
Tesla Ships 497K Vehicles (And Keeps Building Capacity)
Tesla reported Q3 2025 numbers this week. Total deliveries: 497,099 vehicles. Total production: 447,450 vehicles.
Deliveries were up 7% year-over-year from 462,890 in Q3 2024. But production dropped slightly from 469,796 vehicles last year. The gap tells you something about inventory management and demand patterns.
Europe remains weak. Sales slumped partly from consumer backlash against Musk's politics, partly from real competition. Volkswagen and BYD are taking market share with vehicles that simply work better for European buyers.
The US offset Europe's decline. Buyers rushed to purchase EVs ahead of the $7,500 federal tax credit expiration. Ford reported 30% EV sales growth in the same period, hitting 30,600 units. That sounds impressive until you remember Tesla delivered over 450,000 vehicles in North America alone.
Why this matters: Tesla's manufacturing capacity and vertical integration enable autonomous vehicle deployment at scale that competitors can't match. When Full Self-Driving actually works reliably, Tesla can update millions of vehicles through OTA software updates. Instant fleet.
Compare that to Waymo's approach. Custom vehicles. Extensive hardware. Detailed mapping. Controlled rollouts. Waymo delivers roughly 250,000 paid rides per week across five cities. Excellent safety record, proven technology, but limited scalability.
Tesla's betting everything on software leverage. Waymo's betting on hardware certainty and operational excellence. Both strategies have merit. Both face massive execution risk.
The interesting part isn't which wins. It's watching two completely different product philosophies compete with billions of dollars and actual customers at stake.
Design lesson: Vertical integration and manufacturing scale create optionality. Tesla can pivot strategy because they control the entire stack. Waymo's hardware investments lock them into specific approaches. Neither is inherently better, but the strategic flexibility differs dramatically.
Pentagon's Replicator Program Misses Targets
The Pentagon aimed to field thousands of small autonomous drones by late 2025 through its Replicator program. According to WSJ reporting, they missed by a lot.
The program achieved important wins. It reclassified drones under $2,000 as consumable commodities rather than durable property. That bureaucratic change matters enormously. Soldiers can now use drones, lose them, and order more without paperwork nightmares.
The program also streamlined procurement and incentivized rapid innovation from companies like Anduril, Fortem, and Saronic.
But actual deployment? Well below "thousands."
The gap highlights something familiar to anyone who's shipped hardware: prototypes aren't products, and products aren't scaled production. Even with reduced bureaucracy, supply chains constrain everything.
Meanwhile, targeted programs with clear requirements keep delivering. The Army tested Launched Effects autonomous drones at Joint Base Lewis-McChord in August. These systems scout targets, gather intelligence, and conduct strikes autonomously to keep soldiers away from immediate danger.
The Navy awarded contracts in August to Anduril, Northrop Grumman, Boeing, and General Atomics for carrier-based autonomous combat drones. Target price: $15 million per aircraft, half what the Air Force pays for similar capabilities.
What works: specific requirements, willing customers, realistic timelines. What doesn't work: ambitious targets without clear production paths.
The military is learning what every hardware startup learns: shipping is always harder than building.
[WSJ: Replicator Program, Oct 2025]
Defense News: Army Tests, Sept 2025
The Split That Matters
AI augments knowledge work. Physical autonomy replaces physical work.
That's the fundamental difference the Yale study accidentally highlights. While economists track employment in "AI-exposed" occupations like consulting and financial services, autonomous systems are eliminating jobs in trucking, warehousing, and manufacturing.
The trucking industry faces an 80,000 driver shortage projected to double by 2030. Autonomous trucks are already filling that gap. When Kodiak completed 900 autonomous deliveries last year with 100% on-time performance and zero accidents, they proved the business case works.
Amazon operates 1 million warehouse robots. That's not augmentation. That's replacement of specific physical tasks that humans used to do.
The displacement is happening. It's just not showing up in studies focused on ChatGPT's impact on office work.
Next week: construction robots that actually deploy (and why they're all wheeled), the mining operation using fully autonomous trucks that nobody talks about, and whether battery technology will ever give humanoids an 8-hour shift.
Forward this to someone who thinks AI will take their job before a robot does.