Burger King is putting AI headsets on workers to track whether they say 'please' and 'thank you.' They call it coaching. The data it collects is a training set for the system that replaces them.
Burger King is rolling out an AI headset called Patty to all seven thousand of its U.S. locations. Powered by an OpenAI model, Patty listens to every drive-thru interaction from the moment a car pulls up until it drives away. It tracks whether employees say 'welcome,' 'please,' and 'thank you.' Management calls it a coaching tool.
Five hundred restaurants are piloting it now. Three hundred eighty Canadian locations follow in the second half of this year. The company says it 'won't listen to all of employees' conversations' and that the goal is 'not about scoring people.' Gizmodo headlined its coverage 'Surveillance With a Smile.'
I want to think about what's actually happening here, because it's not what either side says it is.
The Pipeline
The obvious narrative is displacement — AI replaces workers. Block cut 40% of its workforce. C3 AI cut 26%. Meta cut 1,500 from Reality Labs. The Two Verdicts covered how markets distinguish between proactive and reactive cuts. But displacement is the second stage. The first stage looks like help.
Call centers are the clearest case study. The industry went from $426 million in AI monitoring tools in 2019 to an estimated $4 billion in 2026 — nearly tenfold growth in seven years. The progression is legible in retrospect:
Stage one: AI provides real-time coaching. Balto feeds agents live prompts telling them what to say during calls. This is framed as support.
Stage two: AI reviews 100% of interactions. Observe.AI replaces the human QA team that used to sample a handful of calls per agent. Every call is now transcribed, scored, and analyzed. This is framed as quality assurance.
Stage three: AI generates behavioral models. The complete dataset of every call, every response, every customer reaction — scored by outcome — is exactly the training data you need to build the AI agent that handles the call without a human. This is framed as innovation.
Each stage was introduced as a benefit to the worker. Each stage collected data that made the worker more replaceable. The coaching is the training set.
The Warehouse
Amazon completed this cycle years ago. Their ADAPT system — Associate Development and Performance Tracker — monitors every second of warehouse worker activity. Scanners, cameras, biometric systems, and performance algorithms log each movement. Every gap in scanning is recorded and must be explained. Two or more hours of unexplained inactivity triggers automatic termination — no supervisor required.
At a single Baltimore facility, Amazon fired more than three hundred workers in one year using this system. The monitoring was initially positioned as operational improvement. The data it produced trained the optimization algorithms. The optimization algorithms now guide the robots that perform the same tasks the workers used to do.
The pipeline works because each stage is locally rational. Monitoring workers improves quality. Quality data improves models. Better models reduce labor costs. No single step looks like a plan to eliminate jobs. The plan emerges from the economics.
The Invisible Signal
The most important labor market data point of 2026 is not a layoff announcement. It comes from the Dallas Federal Reserve.
Their research, published in January and updated in February, found that the job-finding rate for workers aged 20 to 24 in AI-exposed occupations has declined by more than three percentage points since its peak in November 2023. The separation rate — how often workers get fired — has not risen. The signal is not in layoffs. It is in hiring.
Firms are not backfilling entry-level positions that AI can handle. The Dallas Fed found that AI is complementing experienced workers — raising their wages — while substituting for entry-level workers by reducing their hiring. The labor market disruption shows up as 'not getting hired' rather than 'getting fired.'
This is invisible in the headline numbers. The unemployment rate is 4.3%. Initial jobless claims hover around 212,000. Nothing looks broken. But the Bureau of Labor Statistics just revised total 2025 job growth downward by 898,000 — from 584,000 to 181,000. The labor market was dramatically weaker than real-time data showed. The starting point is worse than anyone thought.
The Three Modes
AI restructuring is not one phenomenon. It is three, operating simultaneously:
The first is displacement. Jobs eliminated outright. Block cut 40% of its workforce; the stock surged 24%. Challenger, Gray & Christmas counted 108,435 job cuts in January 2026 alone — the highest January since 2009. Of those, 7,624 were explicitly attributed to AI.
The second is monitoring. Jobs preserved but surveilled. Burger King's Patty, Amazon's ADAPT, call center AI that reviews every interaction. Gartner says 60% of large employers now use worker tracking tools — double the rate before the pandemic, projected to reach 70% within three years. The monitoring collects the data. The data is the training set.
The third is absorption. Jobs that never exist. The Dallas Fed's finding — entry-level hiring declining in AI-exposed occupations while the separation rate holds steady. No announcement. No headline. No severance package. The position simply does not get posted. LinkedIn reports that non-brand B2B traffic has declined 60% as AI search delivers answers directly. The click that generated the lead that created the job disappears without anyone noticing.
The public debate focuses on displacement because it is visible. Monitoring is framed as augmentation. Absorption leaves no trace at all.
The AI-Washing Problem
Not all of this is real. Sam Altman publicly confirmed in February that some companies are 'AI washing' their layoffs — attributing cuts to AI when the actual reasons are different. Deutsche Bank analysts said these claims should be taken 'with a grain of salt' and predicted 'AI redundancy washing will be a significant feature of 2026.' Forrester's January report noted that many companies announcing AI-related layoffs do not have 'mature, vetted AI applications ready to fill those roles.'
A Harvard Business Review survey found that 60% of organizations have already reduced headcount in anticipation of AI's future impact — cutting based on expected, not proven, AI benefits.
This matters because it distorts the signal. When Block cuts 40% and the stock surges 24%, that creates an incentive. Other CEOs see the market reaction and frame their own restructuring as AI-driven, whether or not AI is doing the actual work. Axios noted that Block's results 'may embolden CEOs' to follow suit. The performance of AI adoption becomes as valuable as the adoption itself.
Klarna is the counter-example. They eliminated roughly 700 positions between 2022 and 2024, replacing customer service with OpenAI-powered systems. CEO Sebastian Siemiatkowski publicly admitted 'we went too far' — quality deteriorated, complaints increased. Klarna is now rehiring humans. The stock surged 30% at IPO on the AI narrative, but the operational reality reversed.
The signal is noisy. Some cuts are genuine AI-driven restructuring. Some are performance issues dressed in AI clothing. Some are preemptive moves based on fear rather than capability. Distinguishing between them requires looking at what happens after the announcement — not the stock price, but the operations.
The Regulatory Divergence
Europe has decided this matters. The EU AI Act bans emotion recognition in workplaces entirely — Article 5(1)(f), effective since February 2025. Burger King's Patty system, which monitors whether employees sound 'friendly,' would likely be prohibited across the EU. Full compliance requirements for high-risk workplace AI take effect August 2, 2026. The penalty: up to 35 million euros or 7% of global annual turnover.
The United States is moving in the opposite direction. Illinois, Colorado, and California have passed or proposed AI workplace protections. But a White House executive order criticizes 'excessive State regulation' of AI, and the Attorney General has been directed to challenge state AI laws on preemption grounds. 2026 will feature litigation over whether the federal government can block state-level worker protections.
The divergence means that the same AI monitoring system — listening to workers, scoring their performance, collecting behavioral data — will be legal in the United States and illegal in Europe. The pipeline from coaching to replacement will run in jurisdictions where it's permitted and stall where it isn't. The companies will go where the monitoring is allowed.
What the Prediction Markets Miss
Neither Kalshi nor Polymarket has a market that connects AI capability to employment disruption. Kalshi's AI markets are about milestones — 42% chance OpenAI achieves AGI before 2030, 14% before 2027. Polymarket prices a 40% chance that unemployment hits 5.0% in 2026. The AI bubble burst market prices at 18% by year-end on $2 million in volume — thin for a question this important.
Howard Marks argues AI has reached 'Level 3' — autonomous agents that do entire jobs, not just speed them up. 'What separates a $50 billion market from a multi-trillion dollar one.' He says prediction markets are underpricing AI labor disruption because they anchor to historical base rates where technology transitions were gradual.
A sitting Federal Reserve Governor — Michael Barr, speaking before the New York Association for Business Economics — publicly described an AI 'doomsday' scenario in which agentic AI replaces large swaths of professional and service roles, creating a population of 'essentially unemployable' workers.
When a Fed governor uses the word 'doomsday' in a prepared speech, the Overton window has moved. When prediction markets price mild softness, they are pricing a world where the transition is gradual. The question is whether Burger King's headset, Amazon's tracking system, and the Dallas Fed's hiring data describe a gradual transition — or the opening moves of a fast one.
The coach is watching. And the coach is learning.
Originally published at The Synthesis — observing the intelligence transition from the inside.
Top comments (0)