Divya van Mahajan

AI Companies Are Hiring More

ai-daily-brief-podcast
Listen to episode →

Overview

This episode of the AI Daily Brief (dated 2026-07-02) examines the latest data and discourse around AI’s impact on jobs and employment. The host also covers several headlines including OpenAI’s proposed government equity stake, Meta’s cloud computing pivot, SpaceX’s AI device rumours, and Anthropic’s rolled-back usage monitoring. The central thesis of the main segment is that, while AI is clearly automating some tasks and displacing workers in certain sectors, emerging evidence suggests that companies actively adopting AI are actually growing headcount—challenging simple displacement narratives.

Source video URL: Not provided.

Prerequisites

  • Basic familiarity with the AI industry landscape (OpenAI, Anthropic, Meta, Google, SpaceX/xAI)
  • Understanding of labour market concepts: payroll data, Bureau of Labor Statistics reports, headcount growth
  • Awareness of ongoing public debate around AI and job displacement
  • Familiarity with terms like AI model benchmarking, large language models (LLMs), and AI agents
  • Basic understanding of cloud computing business models (AWS, NeoClouds like CoreWeave)

Main Points

OpenAI’s Proposed 5% Government Equity Stake

  • OpenAI has proposed transferring a 5% equity stake to the U.S. Government, structured similarly to Alaska’s Permanent Fund (oil/mining revenue distributed to citizens).
  • At current valuations, the stake would be worth approximately $42 billion; it is unclear whether this would be a purchase or a gift.
  • OpenAI has proposed that all leading AI developers (Anthropic, Google, Meta, others) also contribute 5% equity—participation from other firms is unconfirmed.
  • The Financial Times characterised the move as a potential quid pro quo to secure favourable relations with the administration and address public concerns about AI wealth distribution.
  • Discussions are described as early-stage and “conceptual,” and implementation may require an act of Congress.

Meta’s Cloud Computing Pivot (“MetaCompute”)

  • Meta is developing a cloud services business called MetaCompute, led by head of infrastructure Santos Janertan alongside Daniel Gross (Meta Superintelligence Labs) and Dina Powell-McCormick.
  • Two potential models: (1) selling access to hosted AI models (similar to AWS Bedrock), and (2) selling raw compute access (similar to NeoClouds like CoreWeave).
  • The announcement caused Meta stock to rise 8.8% on the day (best single-day gain in six months), while CoreWeave and Nebius fell 14% and 17% respectively.
  • Analysts at Jeffries compared the strategy to Amazon’s early AWS pivot; Mizuho characterised it as a “plan B” rather than a near-term revenue driver.
  • This mirrors SpaceX’s cloud pivot, where compute sales to Anthropic, Google, and Reflection AI reportedly became SpaceX’s primary revenue driver.

Anthropic’s Rolled-Back Monitoring in Claude Code

  • A Reddit user revealed that since April, Claude Code had been covertly checking whether users had a proxy enabled and transmitting metadata (including time zone data) to Anthropic to detect usage by Chinese labs or unauthorised resellers.
  • Anthropic’s prior research reports and Congressional letters contained specific usage numbers implying some level of monitoring.
  • Claude Code developer Tariq confirmed the experiment was launched in March and has since been rolled back following the Reddit controversy.
  • The episode highlights the degree of visibility Anthropic has into work performed on their systems.

Fable 5 Re-Release and User Experience

  • Users regained access to Fable 5 (an Anthropic model) on Wednesday to generally enthusiastic reception, with reports of high output quality, speed, and agentic task capability.
  • A significant complaint emerged: many users found that Fable was routing tasks to Opus (a less capable model) rather than completing them itself, with some reporting Opus handling 80% of tokens.
  • Other users reported no refusals. Analyst commentary noted Fable appears particularly strong as an orchestrator of other agents.
  • Professor Ethan Mollick noted that best-practice workflows for long-running agents are still largely unknown—“we are still on day one.”

Remote Labor Index: AI’s Ability to Do Economically Valuable Work

  • The Center for AI Safety updated its Remote Labor Index, which benchmarks models on freelance tasks (3D modeling, graphic design, video/audio editing, data analysis, programming) judged against a gold-standard professional deliverable.
  • GPT-5.5 scored 6.3%; Opus scored 8.3%; Fable scored 16.1%—up from 2.5% eight months prior, representing more than a fourfold increase in under a year.
  • The Center acknowledged that 16% completion of freelance tasks does not equal replacement of 16% of all human jobs; most tasks still fall short of professional quality.
  • Interpretations vary widely: some see an accelerating threat to freelancers; others emphasise that 84% of tasks still require human involvement.

Task Exposure vs. Job Replacement

  • OpenAI’s chief economist Rani Chatterjee argued at a European Central Bank event that task exposure to AI does not equate to job substitution.
  • He used the example of economists in 1985 adopting personal computers: the PC was a complement that increased productivity, not a replacement.
  • Software developers were cited as the most exposed profession, yet significant job shrinkage has not materialised to the predicted extent.
  • Bureau of Labor Statistics data shows tech and finance losing 28,000 jobs/month on average in 2026, while overall hiring adds 113,000/month.
  • John Challenger (Challenger, Gray & Christmas) tracked over 100,000 job-cut announcements this year and noted AI is mentioned at a frequency unprecedented for any prior technology.
  • Analysts at Barclays cautioned that AI is frequently used as a cost-cutting narrative to justify layoffs that may have originated in 2022 overhiring.

China’s Policy Approach: Legally Protecting Workers from AI Displacement

  • China is pressuring companies to avoid AI-driven layoffs; non-compliant firms risk legal action.
  • In April, a Hangzhou court ruled a tech company illegally laid off a worker after replacing him with AI, citing that AI development should “liberate labour” while protecting workers’ rights.
  • The host drew a parallel to a potential U.S. policy approach: providing incentives (subsidies or fines) for firms to retain workers through the “efficiency phase” of AI adoption, accelerating the transition to the “opportunity phase.”

Ford’s Re-Hiring of “Graybeard Engineers”

  • Over three years, Ford hired approximately 300 veteran engineers (“graybeards”), many former employees, to fix persistent quality problems that AI tools were failing to resolve.
  • Ford VP Charles Poon stated that AI is “only as good as the information you use to train it” and that experienced engineers were essential for training AI tools effectively.
  • The result: Ford became the top mainstream brand in the J.D. Power Initial Quality Survey.
  • The host characterises this as a leading indicator of companies recalibrating where AI sits in their labour stack.

RAMP/Revelio Labs Study: AI Adopters Are Hiring More

  • A study correlating firm-level AI spending with payroll data across 21,000 U.S. businesses found that high AI-adopting companies grew headcount by 10% on average over two years, versus flat growth for low adopters.
  • Headcount growth at the entry level was even higher at 12%.
  • Headcount growth began 6–12 months after the start of AI adoption, suggesting a learning curve.
  • The threshold for “high AI adoption” was modest: approximately $30 per employee per month in early phases.
  • Lead economist Eric Karazian noted the study controlled for pre-existing growth tendencies by matching firms against like-for-like control groups; he urged cautious interpretation but highlighted implications for young workers and recent graduates.
  • A Box CEO survey of 1,600+ mid- and large-sized companies found 58% expect headcount to rise over the next three years, rising to 79% among the most mature AI adopters.

Key Concepts

  • Remote Labor Index: A benchmark from the Center for AI Safety measuring AI models’ ability to complete freelance tasks at a quality level acceptable to a paying client, judged against professional gold-standard deliverables.
  • Task vs. Job Distinction: The analytical separation between individual tasks that AI can automate and entire jobs (which comprise many tasks, social functions, and contextual judgements); exposure of a task to AI does not automatically mean the containing job is eliminated.
  • Efficiency Phase vs. Opportunity Phase: The host’s framework distinguishing the initial period of AI-driven cost-cutting and headcount reduction from a subsequent period where AI enables new products, services, and hiring.
  • Tributary Capitalism: A term used to describe the emerging dynamic where tech companies provide financial or equity contributions to government as a condition of operating in a favourable regulatory environment.
  • MetaCompute: Meta’s planned cloud services business that would sell access to hosted AI models and/or raw compute capacity to external customers.
  • Graybeard Engineers: Experienced veteran engineers re-hired by Ford to train younger staff and improve AI tooling, highlighting the irreplaceable value of deep domain expertise.
  • NeoCloud: A cloud infrastructure provider (e.g., CoreWeave, Nebius) that specialises in GPU compute for AI workloads, distinct from hyperscalers like AWS or Azure.
  • Sovereign Wealth Fund (AI context): A proposed government-owned investment vehicle that would hold equity stakes in AI companies for the benefit of citizens, modelled on funds like the Alaska Permanent Fund.
  • Revelio Labs: A workforce data analytics firm whose payroll dataset was used in the RAMP study correlating AI adoption with hiring trends.
  • Distillation (AI): The process of training a smaller or proprietary model using outputs from a more capable frontier model, which AI companies like Anthropic seek to prevent unauthorised third parties from doing.

Summary

The episode argues that the relationship between AI and employment is far more nuanced than either catastrophist or dismissive narratives suggest. Benchmark data from the Center for AI Safety’s Remote Labor Index shows that frontier AI models are improving rapidly in their ability to complete economically valuable freelance tasks—quadrupling in capability in under eight months—yet still complete only 16% of such tasks at professional quality, leaving the vast majority of work requiring human involvement. Macro labour data shows genuine job losses concentrated in tech and finance, but a large-scale study by RAMP and Revelio Labs of 21,000 U.S. businesses found that companies with high AI adoption are growing headcount by 10% or more, with entry-level hiring growing even faster, suggesting that AI adoption correlates with expansion rather than contraction. Anecdotal evidence from Ford’s re-hiring of veteran engineers reinforces the idea that companies which moved quickly to substitute AI for human expertise are now recalibrating. The host concludes that while short-term displacement is real and will affect specific vulnerable populations and job categories, the evidence increasingly supports an augmentation rather than replacement thesis, and that targeted policy support for at-risk workers is a more appropriate response than preparing for a broad “job apocalypse.”