AI at work: corporate psychosis and jobs that won't come back

Today’s most important AI news

The topic dominating this Sunday isn’t a model launch or a new tool: it’s the friction between AI as a corporate mandate and the reality of the people who have to live with that mandate. Three different stories point to the same problem — companies that pressure, employees that pretend and labor markets that are already giving way. Add to that the close of Silicon Valley’s most dramatic trial of the year.

”Use more AI or explain why you don’t”: the corporate psychosis of 2026

A thread by Mitchell Hashimoto — co-founder of HashiCorp — became the most commented post on Hacker News this week, with more than 2,000 points and nearly 1,200 comments. The central argument: entire companies have entered what Hashimoto calls “AI psychosis”, a state in which decisions are no longer made based on real results but on the need to show that AI is being used, at any cost.

The most concrete case comes from Amazon. According to a Fast Company report, the company’s workers are inventing artificial tasks to meet internal AI-adoption metrics. It’s not that the tool doesn’t work — it’s that the KPI exists before the use case. Managers who never defined what they wanted to use AI for now demand that their teams use it a set number of times per week, and employees respond the only way they can: by fabricating activity that looks good on the dashboard.

This pattern has a name in management: Campbell’s Law — when a metric becomes a target, it stops being a good metric. What makes this cycle different is the speed: the pressure arrived before there was operational maturity, before teams knew what problem to solve. For any product manager or team leader designing an AI adoption strategy, the lesson is direct: mandating usage without defining value produces theater, not transformation.

Jobs exposed to AI are already falling, and the data confirms it

Bloomberg published an analysis of the US labor market this week showing concrete and sustained job losses in categories directly exposed to AI automation. It’s not forecasting or modeling — it’s numbers from recent months. The most affected categories include data-processing roles, customer support, document analysis and repetitive administrative tasks.

The coverage on Hacker News generated almost 250 comments, with the usual debate between those who argue that “new jobs are always created” and those who point out that the speed of this transition leaves no time for retraining. What sets this moment apart from earlier automation cycles is that language models don’t need complex integrations or specialized hardware to displace mid-level cognitive work — they lower the cost of entry enough for a small company to automate in weeks what used to require hiring.

For professionals in Latin America, the most immediate risk isn’t mass local unemployment — the region’s labor markets have different dynamics — but downward wage pressure in categories that compete globally for remote work. An analyst in Mexico City or Bogotá doing tasks a model can do today is competing against that model, not against another analyst. The relevant question is which skills make that work irreplaceable.

The Musk vs. Altman trial goes to the jury with accusations on both sides

After three weeks of hearings, the Musk v. Altman case entered its final phase this week. MIT Technology Review’s coverage sums up the last few days as a credibility duel: Musk’s lawyers presented evidence that Altman allegedly lied and operated with undisclosed conflicts of interest between companies doing business with OpenAI. Altman responded on the stand by painting Musk as someone who wanted direct control over the development of artificial intelligence, not simply to protect its original nonprofit mission.

The legal core of the case is still whether OpenAI violated its obligations as a nonprofit organization by becoming a capped-profit company and now an openly commercial corporation. But what the trial has done beyond the law is expose in public records the internal dynamics of OpenAI’s early years: who wanted what, who promised what and how decisions were made that affected billions of dollars and, according to both parties, the course of AI.

The verdict was not yet known at the time of writing. But regardless of the legal outcome, the trial has already had measurable reputational consequences: documents revealed during the proceedings have fueled weeks of critical coverage of OpenAI’s governance. For companies and products that depend on OpenAI’s API, the underlying question isn’t who wins the trial but what governance model emerges on the other side.

In a nutshell

What ties today’s news together is a single tension: the difference between adopting AI because it solves something real and adopting it because there’s external pressure to do so. Amazon fabricating tasks, companies in “psychosis” according to Hashimoto and a labor market that’s starting to feel the impact are symptoms of the same problem — AI advanced faster than organizations’ ability to integrate it with judgment. The Musk-Altman trial adds another layer: not even the field’s founders were clear about where they were heading. That disorder isn’t temporary. It’s the working condition of the coming years.

---

Sources used: https://twitter.com/mitchellh/status/2055380239711457578, https://www.fastcompany.com/91541586/amazon-workers-pressured-to-up-ai-use-extraneous-tasks, https://www.bloomberg.com/news/articles/2026-05-15/us-is-starting-to-see-heavy-job-losses-in-roles-exposed-to-ai, https://www.technologyreview.com/2026/05/15/1137357/musk-v-altman-week-3/