At the end of last year, I tried to do a comprehensive overview of Amazon’s warehousing automation and concluded that we shouldn’t expect fully “dark” warehouses and an associated employment apocalypse from the company, but rather that
we’re likely to see more of the same from Amazon: intense labor surveillance, discipline, and subcontracting/gigification combined with displacement of labor through robotics improvements that are mostly masked through company growth.
Shortly after that, I questioned the narrative that Amazon’s corporate layoffs were really about AI, and since then my skepticism of the “AI/the robots are coming” narrative has only hardened, less because I doubt the transformative nature of the technologies being deployed and more because I don’t think the operational context of AI deployment is being considered in doomer-ish accounts of such things.
Recently, Amazon held an internal meeting to address a
“trend of incidents” in recent months, characterised by a “high blast radius” and “Gen-AI assisted changes” among other factors, according to a briefing note for the meeting seen by the FT. Under “contributing factors” the note included “novel GenAI usage for which best practices and safeguards are not yet fully established”.
This meeting comes on the heels of laying off more than 10% of its corporate workforce in recent months, and new imperatives for all staff to incorporate AI use into their regular work flows. Amazon employees I’ve talked to say that the company is dead serious about having every employee use AI, and documentation of AI use is required in things like promotion packets.
From an individual perspective, this makes total sense. Every software engineer I know thinks Claude Code is a genuine problem for new computer science grads, and that the industry will never look the same. But when they say this, they are thinking about their own productivity. At a company like Amazon, however, it is not simply that individuals need to be productive but that many teams working together to maintain and improve a mind-bogglingly complex system need to be productive, and that is a social and organizational problem, not a technological one.
To be clear, this social and organizational problem can and will be solved, and when it is workers at these big tech companies will indeed be more productive than they were before. But at present, Amazon’s clearly out in front of its skis, and a hiring course correction seems inevitable. (Also: I’m still waiting on a FOIA request at USCIS, but I’m guessing, as with the layoffs of 2022-23, that Amazon onboarded a ton of new H-1B’s, right before the layoff announcements hit in November 2025.)
Independent researcher Vsevolod Shabad has a great short commentary on what’s happening at Amazon, where he worries that the layoffs in combination with rapid AI adoption have created an “organizational amnesia” that it will take a long time to overcome.
Professional judgment is not an innate trait; it is forged through the cognitive friction of entry-level analysis that Amazon is now automating away.
By treating junior coding as a mere inefficiency to be optimised, the company is hollowing out the institutional memory required to recognise a “high blast radius” before it crystallises.
To my mind, there are two options here: either Amazon continues down its current path and risks major incidents that endanger its entire operation, or else it quietly recognizes an excessive enthusiasm for AI adoption and staffs up again. Jassy’s team would really have to be disconnected from the reality on the ground to do the former (very possible, of course!). Meanwhile, the latter, despite all the doom and gloom in the white-collar world, presents quite an organizing opportunity: an existing workforce of skilled workers disillusioned with the company for its callous irresponsibility and newly prized for their institutional knowledge. That seems like a good situation in which to organize!
Either way, the present moment is a “prime” one for Amazon corporate workers to take the lead in defining the parameters of responsible AI deployment, one of the many reasons why white-collar AI doomerism is so pernicious.