ThinkPatternGet the app
Perspective
TECHNOLOGY · NOV 19, 2026

Hours Sold Before They're Worked

Since mid-2025 the AI industry has stopped selling software and started selling labor — but on every ledger that would have to ratify that claim, the hour is still a human's, or not yet worked at all.

Sometime in the past year, the companies that build AI stopped describing what they sell as software. A piece of software is a thing you buy. What they are now describing is work. The most explicit version came from Rotem Alaluf, chief executive of Wand AI, who proposed the "agent-hour" as a unit of AI labor [1], to be counted inside a measure he calls "agentic GDP," meant to count AI labor inside the national accounts — the official ledgers where GDP is measured [1]. Alaluf's framing is blunt.

A workforce you cannot measure is a workforce you cannot manage. — Rotem Alaluf

This is not one executive's op-ed. It is now an industry posture, arrived at from several directions at once. Aisera proposed an Agent Efficiency Index and an "autonomy ratio" tracking tasks completed without human intervention [2]. HPE wants to price computing by dollars-per-FLOP, a productivity measure for infrastructure rather than for software [3]. Anthropic reports that its agents began logging more work hours than its human employees [4]. Robot makers ship "teammates," not machines [5][6]. Every one of them is making the same move: reclassifying the product from software to labor. A reclassification is not a fact. It is a proposal, addressed to counterparties — and it only takes effect when somebody else countersigns it. Three parties would have to sign: reality, labor's own institutions, and the financial system. None of them has. Reality moves fastest, and it has not signed. The "workforce" still runs on human labor. Duckbill, an AI-agent startup, sells an API that lets other developers outsource "atomic tasks" — phone calls, scheduling, CAPTCHAs — to vetted human contractors whenever the agents can't finish the job [7]. In China, startups pay factory workers and cobblers to wear sensor gloves so their physical motions can be recorded and resold as training data for the robots meant to replace them [8]. MIT research found that 95% of disclosed corporate AI initiatives produced no measurable return on investment in the first half of 2025 [9]. And the humanoids: the industry declared its "ChatGPT moment," yet Merrill Lynch projects global shipments of roughly 18,000 units a year — about one mid-sized factory's headcount [10]. The loudest claimant is also the easiest to check. Anthropic's hours figure is unaudited, published ahead of an IPO, and it belongs to a company that posted an $80 billion net loss in the first half of 2026 and carries a junk credit rating [4]. The firm insisting its agents now work more hours than its people is, on its own books, losing money at a scale no workforce of any kind has yet earned back. Labor's institutions run steadier — they have answered with labor's own tools rather than tech policy [11]. Oakland's city workers won a strike-backed three-year deal requiring retraining or reassignment instead of layoffs [11]. Pennsylvania's governor granted SEIU Local 668 oversight of how the state uses AI, and California's governor signed laws banning AI-only discipline and certain AI workplace surveillance [11]. This is contract language and labor law, the machinery unions already know how to run — not a position in a tech-policy debate. The UAW's president, Shawn Fain, gave the stakes a name.

Optimus will be the biggest product ever. — Elon Musk

What the unions are doing is the telling part. They are not debating whether an agent counts as a worker. They are bargaining as if the hours are contested — a new unit of labor proposed on one side, declined on the other. The financial system runs slower and more formally, and its answer is the most explicit refusal yet, delivered in numbers rather than slogans. The Federal Reserve's latest Financial Stability Report names AI the top threat to U.S. financial stability, with the share of surveyed respondents citing it as a salient risk jumping from 50% in May to 87% by October [12]. The buildout is being recorded as debt and risk, not productivity: $680 billion in advanced-computing debt issuance, five times 2023 levels, against more than $3 trillion in total AI-infrastructure debt, guarantees, and leases [12]. Fed Governor Michael Barr warns that agents linked to bank accounts could trigger "agentic bank runs," draining deposits in minutes, and JPMorgan's Jamie Dimon confirms banks are already blocking rogue agents [13]. Hold the quietest entry for last. Analysts find that AI's rapid technical advance remains invisible in real-economy data, with the gap between capability and measurable productivity expected to persist through 2027 [14]. The national accounts this proposal aims at — the very ledgers where "agentic GDP" would be recorded — do not yet contain the product. None of this says the displacement is imaginary. Entry-level marketing shed more than 10,000 jobs in seven months, and Amazon's chief executive says agents drove 30,000 corporate cuts [15][16]. The unions are bargaining against something real. What fails is the reclassification, not the disruption. The hour inside "agent-hour" is, on every one of these books, still a human's or not yet worked at all. The contractor finishing the phone call is the hour. The cobbler wearing the sensor glove is the hour. The factory still waiting on the year's 18,000 robots — its whole headcount in machines — is the hour, deferred. The industry is selling hours of work before the hours are worked.


Sources
  1. 1. Wand AI CEO Proposes Agent-Hour to Measure AI Labor
  2. 2. Aisera Director Proposes New Metrics for Enterprise AI Agents
  3. 3. HPE Leader Proposes $/FLOP Metric for AI Infrastructure
  4. 4. Anthropic Claims AI Agents Reduced Entry-Level Technical Hiring
  5. 5. AGCO to Deploy Agility Robotics Humanoids in Kansas
  6. 6. Automakers Deploy Humanoid Robots Amid Labor Union Opposition
  7. 7. AI Agent Startups Integrate Human Workers for Complex Tasks
  8. 8. China Deploys Embodied AI Humanoids During APEC Summit
  9. 9. Experts Warn Corporate AI Adoption Driven by Herd Mentality
  10. 10. Humanoid Robot Industry Enters Era of Mass Production
  11. 11. US Public Sector Unions Secure Protections Against AI Automation
  12. 12. Federal Reserve Names AI Primary Threat to Financial Stability
  13. 13. Federal Reserve Warns AI Agents Could Trigger Rapid Bank Runs
  14. 14. AI Advancements Show Limited Impact on Real Economy
  15. 15. AI Agents Displace 10,000 Entry-Level US Marketing Jobs
  16. 16. AI Reverses Labor Trends by Displacing White-Collar Workers

Keep reading in the app

The full perspective, free in the app.