This was the week Anthropic took the lead in the AI race. Or at least that’s how it felt.
Seemingly every other tech-related headline referenced potential disruption from Claude-enabled tools. The market fallout was dramatic and far-reaching, even as the most acute pain manifested in sector-specific routs.
The most well-publicized casualty were software shares. It took a sharp Friday rebound for the associated ETF to avoid its worst week since the financial crisis.
Even with the late-week bounce, the product was still down more than 9%. On a four-week rolling basis, a window which captures the mid-January launch of Anthropic’s Cowork, the drop counted among the largest on record for the vehicle.
“SaaSpocalypse” was already in the mainstream market lexicon when Anthropic rolled out the latest version of its flagship offering on February 5. “We’re upgrading our smartest model,” a post announcing the release began.
Opus 4.6, the upgrade, can perform “a range of everyday work tasks,” Anthropic said. Those tasks include “running financial analyses [and] doing research.”
That was just about the last thing shares of FactSet, which were already on the back foot, needed to hear. The stock fell nearly 8% Thursday, just a day removed from a 10% plunge.
As the figure shows, the two-week selloff for FactSet may as well be the largest ever.
Some of this selling is doubtlessly overdone, and I’m an avowed skeptic regarding some of the more apocalyptic “human replacement” worries surrounding AI. That said, we shouldn’t delude ourselves: When it comes to repetitive tasks and number-crunching, these models will render large numbers of people redundant.
One of the more alarming aspects of Cowork — the agentic AI tool mentioned above which started the SaaS panic last month and which, via an extension unveiled this week, can now automate some legal tasks — is the fact that it built itself. In just 10 days.
Very long story short, Anthropic realized its best-in-class AI coding tool might be the key to sprinting ahead in the overall race for AI supremacy. All the company had to do was ask Claude Code to build a suite of agentic tools.
As Boris Cherny, Claude Code’s creator, put it, explaining the impetus for Cowork, “We saw people using [Claude Code] for all sorts of non-coding” tasks. Anthropic took that and ran with it. They told Claude Code to create a product that assists people in their everyday work unrelated to coding. And Claude Code delivered that product in less than a week.
In the post announcing Opus 4.6, Anthropic said that in addition to running financial analysis, the upgrade can “also apply its improved abilities” to “creat[e] documents, spreadsheets and presentations” inside Cowork, where Claude “can multitask autonomously.”
The figure above — hat tip to Charlie McElligott — shows the ratio of the software ETF to the most popular semiconductor product. It speaks for itself.
The irony in all of this is that Anthropic’s supposed to be the benevolent AI startup. As Parmy Olson wrote for Bloomberg Opinion this week, protecting humanity from the existential risk posed by AI is “a kind of ideology” for CEO Dario Amodei.
Less than a year ago, Amodei beseeched his fellow AI developers, as well as politicians, not to “sugar coat” the inevitable. Summarizing an interview with Amodei last May, Axios said he predicted AI will eliminate “half of all entry-level white-collar jobs and spike unemployment to 10-20%” amid a potential mass extinction event for positions “across technology, finance, law [and] consulting.”
Data released this week showed job openings for those positions fell in December to the lowest since 2017 excluding the months around the onset of the pandemic. “The safety of employment doesn’t seem to factor into Anthropic’s ideals,” Olson dryly quipped of Amodei’s pretensions to “align[ing] AI with human values.”
At the risk of trafficking in the same sorts of AI doomsday narratives I don’t much care for, I can’t resist: Ask not for whom obsolescence looms; it looms for thee.
I imagine that many of my prompts were part of the data analysis that led to the Cowork product. I was using Claude Code not only to build and validate software products but also to perform cited market research about the target market and penetration opportunities. This was way back when Opus 4.1 was the premium model, 5 months ago.
An interesting aside to this whole Anthropic story, they actually seem to be cognizant of the harms that Generative AI can cause and are actively trying to build models that avoid problematic behavior. I appreciate that effort and intentionality as much as I appreciated Google’s don’t be evil mantra. (Until they decided to sell us all out for ad revenue)
The sad part about this whole story is that as a retail investor, I have no path to invest in this company. Private equity will reap the rewards of the AI firm that realized what their mission was early and executed on it intentionally. My gut instinct says they probably already have Opus 5.0 ready to go and are just waiting for the competition to drop their next model before they drip out the next iteration Claude improvement that slightly beats them. It’s been effective so far.
Either way, I can’t praise what these models have enabled me to accomplish over the past 6 months, truly life changing technology.
I sent the link for Claude/Cowork to my daughter.
🙂
I’m already doing the white-collar equivalent of doomsday prepping: chasing meme stocks hoping that I hit it big enough to secure the future of me and my children but keeping a Wendy’s application handy just in case.
Seriously though, if you don’t think AI is coming for your job and respond accordingly, you are going to get wrecked.
What is ‘respond accordingly?’ Last I checked, there’s no playbook for this.
Either figure out how to use AI in your current job and stay on top of the latest developments or find an industry where AI won’t double (or more) people’s productivity.
I have one person on my team now because I can do by myself what I would’ve needed two or three analysts to do a few years ago and that’s just using ChatGPT and Gamma.
It must suck to be in high school now, trying to figure out how you’ll make a living. More than technology, it’s the pace of change. Traditional white collar jobs will certainly take a beating and not everyone can or wants to be an electrician or plumber. Even as a college freshman, your chosen field will probably change dramatically at least 3 times before you graduate. Of course there will still be gaming, prediction betting, AR porn and THC. Gotta keep the masses asleep so they don’t remember where they left the pitchforks and torches. Feels like we are heading for multiple tipping points all at once. Perhaps the biggest one – climate change – is all but forgotten with the daily drama circus our news has become.
“Of course there will still be gaming, prediction betting, AR porn and THC.”
This is probably a good time for everyone to revisit “Dope” and “Hustle”…
https://heisenbergreport.com/2025/05/03/dope/
https://heisenbergreport.com/2025/07/02/hustle/
Example #1 – There are so many wonderful things to bet on during the “Big Game” (I don’t want the NFL to sue me for saying Super Bowl.) My favorites include over-under wagers on the length of the national anthem.
It does remind me of the waning days of the Roman Empire …
There will always be work for economists. It’s hard to get fired when it’s impossible to be right. One might try to say the same about meteorologists, but unlike economists, who merely pretend to be good at math, to be a meteorologist you have to actually be good at math.
Source: my economics degree
We just went through a week of sleet, snow, freezing rain and high winds here. The forecasting was excellent and they kept mentioning they were using AI. Impressive or very lucky.
There will always be work for engineers. I find AI helps me do more quality control and deep research. It also helps dme navigate government regulations. What a tool to cut through government regulations? All hail government regulations!!!! However it does not eliminate risk nor develop judgement.
hale
Hal 🙂
Question for the group regarding Opus/Claude and its progeny (more to come).
What hardware must it run on? Does it require the power of the massive computer database campuses being built, many/most powered by NVDA chips, or can some/most of it be run more locally?
I ask in all earnestness, because I have no clue.
Thanks
The current models run on servers which are much more powerful than your home computer. More powerful than most corporations run their operations on. Therefore people access these tools using the internet. Innovation will whittle away and maybe even eliminate that requirement, but for now that is a pipe dream.
Thank you.
I have to disagree with some of the other comments. Engineers, economists, mathematicians, meteorologists, and unfortunately even financial/political writers will be displaced. The only jobs that will be left will be those that need hands, fingers and backs. And not for typing. Physical jobs, the ones considered menial today.
This writer ain’t goin’ nowhere, or at least not on account of any robot. I can assure you of that.
On a related note: One thing we’re overestimating is the appetite for AI-created art. AI can produce new Doors albums, new Beatles singles and Elvis-Tupac collaborations all day, but would anyone seriously want to listen to that beyond the novelty of the first spin or two? The same with literature. I’m sure ChatGPT can write “Crime and Punishment II: Raskolnikov’s Redemption” or “The Sun Also Rises, Again,” but nobody’s going to read them.
I don’t know…if AI redid the last few seasons of Game of Thrones or LOST, I’d watch.
AI may not completely replace artistic expression, but you’re underestimating how many people enjoy the schlep that Netflix produces.
I’m glad you’re sticking around though. Does that mean we can assume your contingency plan for life after Heisenberg doesn’t involve a Black Mirror version of yourself?
Dang it, got my Yiddish words mixed up. I meant dreck.
My “near term” AI fears are more what I guess one could call “macro”: the potential for havoc on the 2026 US mid-term elections. The AI agent ability to create and enhance narratives is immense.
I’m curious how this Anthropic moment will ultimately compare and contrast with last year’s Deep Seek episode…