Five Things AI: Anthropic, Bad Morale, Little Productivity Gains, No Firing in China, Orchestration Tax
There is so much AI out there! Find out what it does!
Heya and welcome back to Five Things AI!
Anthropic just became the most valuable AI company on the planet, and the gap between the builders who take safety seriously and the ones who act like Marvel villains keeps widening. This week also brought layoffs gutting tech worker morale, a reminder that individual productivity gains from AI are not yet showing up in the macro numbers, China telling its companies to grow with AI rather than shrink headcount with it, and a sharp piece on the hidden cost of agentic workflows that hit close to home. On that last note: I shipped klaar (give me a star on Github!) today, a local Rust tool for working with AI agents more precisely, and the orchestration tax article explains exactly why something like it needs to exist.
Anthropic Tops OpenAI to Become the World’s Most Valuable A.I. Start-Up
On Thursday, Anthropic punctuated its ascent by officially passing OpenAI as the world’s highest flying A.I. start-up. Anthropic said it had raised $65 billion in financing that values it at $900 billion before the inclusion of the new capital, a deal that puts it ahead of OpenAI’s last valuation of $730 billion. The company also unveiled a new flagship A.I. model, Claude Opus 4.8, which is significantly better than its predecessor at generating computer code.
I don’t want to say “told you so!”, but it was very obvious this would happen. Sam Altman was too focused on world-domination and too distracted by other projects and ideas that Anthropic paced ahead. Their models feel smarter and more human than ChatGPT, while at the same time Dario Amodei behaves much more like a sane person than Sam Altman does, who comes across like Elon Musk lite.
The morale of tech workers is plunging as layoffs mount
The shift has had practical effects. A Meta employee said in an interview that some workers on her team now used less vacation time and that, in a break with custom, people frequently checked on their projects while on vacation. They increasingly worry about getting a poor performance review or losing their job if they aren’t constantly available.
The employee, who declined to be identified for fear of retribution, said she and many of her colleagues frequently checked Blind because it could be comforting to see how many other Meta workers shared their anxieties.
Employees at several companies said in interviews that their morale was further undermined by the feeling that the layoffs were abrupt and arbitrary, and executed with little empathy.
Reading the news about rounds of layoffs at Big Tech from afar really feels like humans are tossed to the curb whenever the stock performance demands it. It’s a no-brainer that this is not improving morale.
Employees using AI are working faster, but the economy isn’t more efficient. A look at what happened in the pre-Internet era might explain why
Labor productivity has seen solid gains in recent years, but TFP has struggled to post significant growth since a post-pandemic surge. The Fed researchers interpreted the divergence as employees working faster and more productively on an individual level, but the workforce as a whole hasn’t necessarily become more efficient.
This pattern mirrors what happened during the computer and internet boom of the 1990s. Starting around mid-1996, labor productivity began accelerating more rapidly than TFP, but the full productivity benefits of the Internet didn’t materialize in the overall data until several years later.
The Nobel laureate Robert Solow encapsulated the dissonance with a quip that has since been immortalized: “You can see the computer age everywhere but in the productivity statistics,” he wrote in 1987.
I think that companies are still fundamentally struggling with AI and what it means for them. It’s all happening to fast and so they settle for Co-Pilot, which of course doesn’t help much in terms of productivity. Small companies can adapt faster than large enterprises where the IT department tends to be hesitant to deploy new software tools.
China Wants Its Companies to Embrace AI—Without Firing Workers
China’s campaign to accelerate AI adoption, released last August and dubbed “AI+,” gives priority to using the technology in sectors such as manufacturing and logistics, which aren’t as sensitive to white-collar job displacement.
Late last year, China’s Ministry of Human Resources and Social Security told employers, particularly tech companies with younger workforces, to refrain from firing employees as they embrace AI, people familiar with the matter said. While both state-owned and private companies have long had to secure a signoff from regulators before conducting large-scale layoffs in China, employers are now being asked to explain layoffs—and, in some cases, to prove that the cuts aren’t because of AI replacing jobs, these people said.
Recent publicized labor disputes in China have sent a similar message, prompting companies to think twice before firing a worker due to automation.
I think the Chinese approach is better for employee morale and will also lead to more productivity when AI is not used to cut costs but to increase turnout.
The Orchestration Tax is You
There is this hidden asymmetry in agentic workflows. Starting an agent is very cheap. It is just a keystroke or a sentence prompt. But closing the loop on the agent is not cheap at all. Someone has to check if what came back is correct and reconcile it with whatever the other agents touched. That someone is you. And there is exactly one of you.
This is so important. I try to be better at this, but I oftentimes start something new while I am waiting for an agent to finish and then I do something else and then I forgot what I was doing in the first place.
