It won’t be painless. The AI revolution, I mean.
I can’t say I’ve consulted the S&P North American software index until today, but I reckon I’ll need to get in the habit, at least for a spell — because it’s absolutely beset, and the proximate cause is AI disruption.
On Tuesday, the gauge was down 5% through midday for a sixth consecutive decline. For 2026, it’s off a harrowing 20%. I’ll generously call that an inauspicious start to the year.
As the figure shows, this has been a one-way ticket lower since late-October, when things started moving in the wrong direction. In recent days, the losses snowballed.
Stripped to the proverbial studs, this isn’t an especially complicated story. There’s a threshold beyond which prospective disruption makes it impossible to value businesses. Humans being emotionally unstable creatures, we start to panic long before that threshold’s actually reached, let alone breached.
Plainly, agentic AI poses a significant threat to SaaS businesses. You don’t need to be any sort of expert to know that. Whether the threat’s overstated is for now impossible to say. My guess is yes, it’s overstated, if perhaps not by a lot. Market participants are describing this in existential terms. Existential framing’s almost always an overstatement, and thank God for that.
The figure above shows you the drawdown for application software and system software stocks. It’s nearly 30%. You’ll note that’s not nearly as deep as the 2022 bear market.
The locus of concern is Anthropic’s Cowork, which further destabilized an already tenuous situation in software shares upon launch last month. Anthropic calls the tool “the future of AI at work.”
Claude “doesn’t just answer questions” anymore, the company bragged. “But actually does the work alongside you.” That sounds great, but the promise is only as good as the work. And there’s already some debate about that.
“Cowork has an alluring promise: That it can apply all the power of Anthropic’s coding agent, Claude Code, to nonprogramming tasks in a simple tool,” The Information wrote, in a review of Cowork’s functionality. “The reality is, like many freshly hired human co-workers, Anthropic’s tool has some potential — if you’re willing to invest the time.”
Most people won’t be. Willing to invest the time, I mean, which suggests the “sell everything regardless of price” attitude among investors may be an example of emotion trumping reason.
I’m no expert, but I do regularly test AI’s capabilities on just the sort of tasks Anthropic’s advertising for Cowork. I’d say two things. First, there’s nothing “new” about being able to coax an AI tool into, for example, extracting a .csv file from a screenshot of a chart. Second, and sticking with that example, there are software solutions that can perform that task more quickly and more efficiently than the latest versions of “brand name” AI models.
Panning out to a 30,000-foot view, the selloff in software stocks tied to the threat posed by agentic AI demonstrates that markets really do “hate uncertainty,” as the saying goes. Some wondered recently whether that old adage still holds given equities’ alleged obliviousness to geopolitical upheaval.
The nature of the (unanswered) question is indeed existential. As one analyst put it Tuesday, “The draconian view is that software will be the next print media or department stores.” Maybe such a dire outlook will come to pass, but remember: Print media and department stores didn’t die overnight. Hell, Saks didn’t file for bankruptcy until last month.
According to the above-mentioned piece published in The Information, Cowork’s ok if you have “some facility in coding,” but even there it’s “a bumpy experience.” It’s difficult, the article went on, “to imagine the product taking off with a general audience until it offers a gentler learning curve.”
Investors don’t care about the nuance right now. They’ve hate-sold software stocks to such a degree that a Morgan Stanley gauge trades — get this — 37 turns cheaper than the decade average.
Bloomberg quoted a Jefferies trader named, appropriately, Jeffrey. “We call it the ‘SaaSpocalypse,'” he said. Somebody get Jeff a drink. And a Xanax. And a straw — for the Xanax, not the drink.
“According to the above-mentioned piece published in The Information, Cowork’s ok if you have “some facility in coding,” but even there it’s “a bumpy experience.” It’s difficult, the article went on, “to imagine the product taking off with a general audience until it offers a gentler learning curve.”
Thanks for that bit of common sense.
Maybe companies will be forced to hire back some junior coders. Or force companies which relied on packaged software rather than develop it in house to hire some coders. Which might seem to contradict the promise that AI will reduce your costs.
This is a true story: my daughter works in finance for an international public company. Last year, the company spent about $1M on “AI based” software to provide some detailed financial analysis that was being requested by upper management.
Nine months and $1.2M later, the AI software didn’t deliver.
My daughter excels at Excel. She weirdly loves using Excel, and is adept at utilizing the more complicated functions for data sets, calculations, etc. She spent her spare time at work, over a 3 week period, developing an Excel spreadsheet that provided the requested info.
At this point, the AI project has been “shelved” in lieu of an Excel spreadsheet. I guess “garbage in, garbage out” can also be applied to AI software.
thanks for sharing, these real world stories are very enlightening
More importantly, I hope that she got the credit that she deserved. Good on her.
She doesn’t want a career- she is super capable, but just wants to stay home (she declined to go back to the office, in lieu of working from home) and have children. She doesn’t want a promotion and can’t wait to leave her job. Unfortunately, she will need money. These 2 things don’t “mesh”.
I’m just watching her navigate her life (not how I did it, but that is probably a good thing), but I might make a “low-key” suggestion that she learn Claude (she’s smart and could do it) and set up her own consulting business in her industry. She has a lot of connections.
Certainly that is true: garbage in, garbage out. What I’m seeing a lof of is that companies have a strong sense that they need to do something with AI, but they’re not sure what. A lot of leadership is then abdicating the decisions the same way they always have: delegate to IT, hire a consultant, buy some packaged software ‘solution.’
AI will reflect whatever you give it and ask of it, so it’s on the user to learn how to apply strategic thinking and use AI to supplement what they already do faster, smarter or more efficiently.
AI will be done in by a lack of applications capable of making money. Right now the money is mostly being made in infrastructure. It will be more useful when a couple of huge “killer apps appear. Chat bots aren’t killer apps They are just annoying.
The old consulting saw that you can get your solution fast, cheap or correct, but are limited to just two of those three options does not apply in this case. She delivered faster, cheaper and “correcter.” This consultant (me) hopes she got/gets a $1+ million bonus and/or some short vesting options as an exempt employee.
Our finance ops teams leverage a lot of Excel macros and formulas to automate things AI can do. They work great until you try to scale them, now you effectively have what should be a database driven solution living in a fragile spreadsheet that has to be carefully managed and protected.
What would probably be more effective is to build an algorithm or stack of algorithms that can execute the same functionality with load balancing and database replicas that runs on a schedule to produce the reporting requested without a human having to do the manual labor.
But I agree with your point, AI is ubiquitous and in places where it really can’t execute well.
As someone who works in the Claude app almost daily, I think you probably need to spend a little more time with Cowork. Yes, it can parse a CSV file, or 15 of them. It can also draw conclusions about that data, construct a new .XLSX with formulas applied to provide you with a fully baked spreadsheet to send out. It can parse all of your documents, draw conclusions and then generate a PPTX for you to present. A new thing it’s started doing is generating interactive web pages that tell a much better story than a slide deck can and really wow your audience.
Outside of Cowork, Claude Code is able to build you a custom SaaS solution within days that has all of the integrations you need, all of the access controls you want, and is capable of being hardened to the point of passing all security checks. It can then integrate with whatever hosting provider you choose and manage your deployment for you. Why would a company pay a SaaS firm to sell them services that never meet SLA’s and don’t natively integrate with all of their services when they can hire a couple of mid-level devs with Anthropic API accounts instead?
That’s the real threat right there.
And the real opportunity
In Q3 it was OpenAI scooping Anthropic; in Q4 it was Google scooping everyone; now it’s Anthropic scooping the economy. When will we learn that these people have mastered PR (using their own tools which master language) and that the proof of the pudding is in the eating, to quote a wise man I read over the weekend?
Not that AI isn’t disruptive, not that it isn’t transforming the very nature of the work I do and forcing me to innovate, when I’d rather coast into an early retirement. But they can’t ALL be game-changers, these press releases. Sooner or later we need to wise up!
(This month’s experiment is, in fact, to use Claude as much as possible for my work. It’s not going well for Claude qua Claude, but the context-engineering magic that Claude’s fans have built up around it is proving beautifully novel and useful AND it works with other vendors’ models. The human factor has once again prevailed in the productivity battle, where AI is so far an enabler…)