Effortless for Whom?

Amazon’s shopping interface turns an enormous operating system into a button and a date.

The customer sees an object, a price, a purchase control and an arrival estimate. The network required to make that estimate credible—inventory placement, software, fulfillment centers, transport, contractors and workers—recedes behind the screen.

In December 2024, the majority staff of the U.S. Senate Committee on Health, Education, Labor and Pensions published a 160-page report after an eighteen-month investigation of Amazon warehouse safety. The committee said it received stories from nearly 500 workers, conducted 135 interviews, and reviewed more than 1,400 worker-supplied documents, photographs and videos. It concluded that speed and productivity requirements contributed to hazardous conditions and reported that Amazon warehouses recorded more than 30 percent more injuries than the warehousing-industry average in 2023.

Amazon rejected the report’s premise, methods and comparisons. The company said it had increased delivery speed while reducing injuries; reported a 28 percent reduction in its U.S. recordable incident rate and a 75 percent reduction in its lost-time incident rate from 2019 through 2023; argued that faster delivery came from placing products closer to customers and reducing process steps; and pointed to a Washington-state decision vacating ergonomics citations. Amazon also described ergonomic redesigns, voluntary worker feedback mechanisms and safety programs.

Amazon has since published later safety updates reporting further reductions in its recordable and lost-time incident rates through 2024 and 2025 and describing continued investment in ergonomics, robotics and safety programs. Those figures are corporate-reported and are not independently audited in this article. They nevertheless belong in the visible record because they postdate the Senate report and are material to Amazon’s continuing account of redesign and improvement.[3]

Both records matter. The Senate report is a majority-staff investigation, not a court judgment. Amazon’s response is the company’s own account, not independent adjudication. Days after the report, OSHA and Amazon fully executed a corporate-wide settlement covering ten Occupational Safety and Health Review Commission dockets. The agreement amended one Illinois citation to a low-gravity serious violation with a $145,000 penalty and withdrew the cited items in the other nine matters. It expressly says the agreement is not intended as Amazon’s admission of the citation allegations. It also requires a corporate ergonomics program with risk assessments, training, employee-reporting routes, program evaluation, biannual meetings, and OSHA monitoring rights during its term.[5][6]

This record does not permit a slogan in place of analysis. It does permit a better question:

Design culture often treats friction as a defect. Extra steps can block access, waste time and cause abandonment. Removing unnecessary work can be an unambiguous good: a shorter form, a clearer instruction, an accessible payment process, a better-routed package.

But “frictionless” usually describes the experience from a chosen vantage point.

The customer no longer searches through a warehouse; a ranking and inventory system must locate the item. The customer waits less; the network must coordinate more tightly. The interface appears simple; exceptions are handled elsewhere. Some of that improvement comes from genuine redesign rather than harder labor. Amazon’s response makes exactly this contrary case: speed can increase because products move closer to customers and the network requires fewer steps.

That is why friction transfer should not be used as an accusation whenever a service becomes easier. It is an investigative proposition. For each removed demand, ask whether the demand was eliminated through better design, automated, absorbed by the institution, or shifted onto a worker, contractor, creator, customer or community with less control over the promise.

The 2024 IRS Direct File pilot supplies a useful cross-sector limiting case. It used interview-style questions to guide eligible taxpayers with simple returns through filing directly with the IRS at no charge. GAO found that the IRS followed leading pilot practices and that taxpayers reported Direct File was easier than methods they had previously used. Treasury reported 140,803 accepted returns and estimated $5.6 million in avoided preparation fees, while disclosing $10.5 million in foundational product costs and $2.4 million in operational costs, excluding U.S. Digital Service costs.[7][8]

That case does not prove the effort disappeared: the IRS and USDS built and operated the service, support staff handled questions, eligibility was limited, and the program later ended. It shows a legitimate form of institutional absorption. Complexity can move away from the user because the institution takes responsibility for simplifying and supporting the route, not because a less powerful worker is necessarily pushed faster. That is why the taxonomy needs both “institution-absorbed” and “shifted.”

Abstraction is necessary. A useful purchase screen cannot display an entire supply chain. A booking page cannot narrate every contingency. A generation button cannot expose every technical process behind a computation.

The issue is not invisibility alone. The issue is decision-relevant invisibility.

Would showing the demand change the promised deadline, staffing decision, price, safety control or customer expectation? Does the institution possess the information while the person carrying the demand lacks the authority to change it? Is the invisible work measured only when it delays the recipient, not when it burdens the producer?

In the Amazon record, those questions are contested with unusual clarity. The HELP majority report points to rate, takt time and time-off-task tracking; repetition and workstation design; and internal projects concerning speed and injury. Amazon says its performance expectations are reasonable, based on what teams at a site are accomplishing, and visible to workers; it rejects the report’s interpretation of internal projects and says ergonomic changes have improved conditions.

The disagreement is not a reason to omit either side. It shows why the design promise and the production conditions must be analyzed together.

NSAG’s working taxonomy identifies five forms of friction transfer. These are analytic categories, not findings that every service produces all five.

Temporal friction moves waiting or slack from the recipient into compressed deadlines, irregular schedules or narrower recovery margins elsewhere.

Cognitive friction moves remembering, coordinating, interpreting and exception-handling to workers, contractors, creators or automated systems.

Relational friction moves the work of calm, explanation, responsiveness and conflict absorption onto the person who faces the customer or community.

Bodily friction moves convenience into lifting, reaching, walking, driving, repetition or exposure. OSHA’s 2023 Amazon citations described risks associated with high-frequency lifting, heavy items, awkward postures and long hours; Amazon contested the cases. The 2024 settlement then established company-wide ergonomics obligations and monitoring.

Reputational friction moves responsibility for a system’s failure onto the most visible person, often someone who did not set the promise and lacks authority to repair it.

The categories can overlap. They should not be casually applied to individuals without evidence. Their purpose is to force the map beyond the recipient-facing interface.

The governance problem sharpens when authority and burden separate.

An organization sets a delivery estimate; a local operation must fulfill it. A platform establishes a service level; a frontline person handles the exception. A brand promises “effortless”; workers must perform the coordination that keeps difficulty outside the customer’s view.

The actor making the promise may capture the commercial value. The actor carrying its failure may absorb the deadline, explanation, bodily demand or reputational consequence.

This is not unique to Amazon, and this essay does not use one company to prove a universal labor theory. Amazon is useful precisely because the public record contains a detailed government investigation, a detailed corporate rebuttal, prior OSHA enforcement and a negotiated corporate-wide remedy. Together they let us see the governing problem rather than merely admire the interface.

The alternative to frictionless design is not deliberate inconvenience.

Some friction protects agency and safety: a confirmation before an irreversible payment, time to check a dangerous exception, review before publishing sensitive information, a humane pace for physical work.

Some friction is extractive: cancellation mazes, inaccessible forms, repetitive proof after an institution loses information, or obstacles that make a right technically present but practically unusable.

Some effort properly belongs with the institution because the institution created the complexity. Some can be removed through engineering. Some should remain visible because hiding it conceals material risk. Some should be shared with the recipient because no ethical system can promise unlimited immediacy at zero cost to everyone else.

The better standard is not maximum friction or minimum friction. It is responsibly allocated friction.

That standard asks:

The button can become faster. The service can become calmer. The network can become more efficient. Those may be real improvements.

But “seamless” describes one position in a system. It does not describe the system.

The honest design question remains: effortless for whom? This question runs through the whole Nervous System Aware Governance series, including Experience Is Governance and Who Gets a Protected Nervous System?