What Wall Street Gets Wrong About SaaS

Last week, B2B software companies tumbled in the stock market, dropping over 10%. Software stocks have been trending down since September 2025, now down 30% according to the IGV software index. The prevailing sentiment is because AI tools like Anthropic’s Claude are now capable of doing things companies used to pay thousands of dollars for.

Chip Cutter and Sebastian Herrara, writing in the Wall Street Journal:

The immediate catalyst for this week’s selloff was the release of new capabilities for Anthropic’s Claude Cowork, an AI assistant that lets users assign agents to perform many types of tasks on their computers using only natural-language prompts. The tools automate workflows and perform tasks across a gamut of job functions with little human input.

The new plug-ins released about a week ago can review legal contracts and perform other industry-specific functions. An update to its model Thursday enhanced capabilities for financial analysis. 

Enterprise software stock vs S&P 500, Aug 7, 2025 – Feb 6, 2026, Source data: Yahoo Finance

But investors are wrong. Nvidia CEO Jensen Huang calls this line of thinking “illogical.” Mark Murphy, an analyst at JPMorgan uses the same word to describe the “expectation that every company will hereby write and maintain a bespoke product to replace every layer of mission-critical enterprise software they have ever deployed.” 

Indeed, the market’s overreaction ignores generations of evidence that proves just because a new technology enables companies and consumers to DIY something, doesn’t mean they will.

Remember when Zapier launched? It heralded the beginning of the no-code movement. Complex and costly integrations between applications was now easy enough for a non-developer to assemble. Systems integrators were going to be in trouble.

The Integrations Revolution That Wasn’t

Zapier was founded in 2011 because its creators kept getting the same freelance request: connect this app to that app. Along with IFTTT and a wave of similar tools, it made a bold pitch—non-technical people could wire their own software together, no developers required. Zapier scaled to 600,000 customers by 2015. Gartner coined the term “citizen integrator.” Pundits declared these tools were “ready to kill software giants.” The writing was on the wall—or so it seemed.

It didn’t play out that way. The no-code tools handled simple, linear automations brilliantly—syncing a form to a spreadsheet, cross-posting to social channels, firing off notifications. But anything beyond that hit what the industry quietly calls the “no-code ceiling.” Non-developers could do the basics, but the moment an integration needed error handling, security, or governance across multiple systems, someone with coding experience was still needed. Teams stacked workarounds and custom scripts on top of their no-code flows, eroding the original promise of simplicity. The tools made simple things simpler. They didn’t make hard things easy.

Worse, the self-serve tools actually accelerated the adoption of the very thing that creates integration demand: more SaaS apps. According to MuleSoft’s 2025 Connectivity Benchmark Report, the average organization now uses 897 applications, with only 29% of them integrated. And every new SaaS tool, cloud migration, or AI deployment creates fresh integration surface area that a trigger-action automation can’t touch.

So what actually happened to the systems integration market? It didn’t shrink. It exploded. The global system integration market hit $410 billion in 2024 and is projected to more than double to $932 billion by 2032—during the exact period no-code tools were supposed to be eating it alive. To be fair, the self-serve tools did nibble at the edges. The big integrators lost a point of market share over five years. But in absolute terms, their revenue kept climbing. The pie grew faster than the slice shrank. It’s the classic pattern: tools that make simple things easier accelerate adoption of the ecosystem, which creates exponentially more complex problems at the next layer up. The systems integrators didn’t lose their jobs. They got a lot busier.

From Million Dollar Websites to $39 per Month

In 2001, I worked on Sega.com. It was the peak of the dot-com boom. I worked at the largest digital agency in the world, marchFIRST. And I was leading design on a $1,000,000 website. To be sure, it wasn’t an easy brochureware marketing site. It was marketing and e-commerce, and had to work on desktop browsers and the Dreamcast console browser on a TV.

Fast-forward to today and one can get a basic Shopify website for $39 per month. There’s clearly decline in revenue for designers, right?

Not even close. marchFIRST went bankrupt in April 2001—the largest digital agency in the world, dissolved barely a year after it formed. The whole dot-com web design economy seemed to be collapsing around it. Squarespace launched three years later. Wix followed in 2006. WordPress was already making simple sites free. The “web designers are dead” narrative has been running for two decades now, through every wave of drag-and-drop builders, pre-made templates, and now AI-generated layouts. And through all of it, the U.S. web design services market has grown to an estimated $47.4 billion.

Think about the scale of that shift. In 2001, there were roughly 29 million websites on the entire internet and the Bureau of Labor Statistics didn’t even track “web designer” as an occupation. Today there are over a billion websites, and the BLS counts over 200,000 web designers and developers in the U.S. alone. The tools got radically cheaper. The profession got radically bigger.

But the Shopify story is where this really mirrors iPaaS. Yes, you can launch a store for $39 a month. 5.6 million merchants have. And yet the ecosystem of developers, designers, and partners servicing those merchants now generates $12.5 billion in revenue—building the custom themes, integrations, and functionality that a $39 plan can’t provide. The platform didn’t eliminate the need for professional web work. It created millions of new customers who eventually needed it.

Same pattern, longer timeline. The unit price collapsed—from a million dollars to thirty-nine—but the number of people who needed a web presence went from a few hundred thousand businesses with the budget for a custom build, to essentially every business on Earth. Every wave was supposed to finally finish off professional web design. Every wave expanded the market instead.

What AI Really Brings

I was catching up with my friend Michael Chou the other day. He’s a product leader and is now heading up product at LumaHealth, an operational AI platform for healthcare. Reacting to the recent market stumbles of software stocks, he had a measured reaction. Despite the billions of dollars being invested in AI and building out AI infrastructure, he likens this AI bubble to that of the push to the cloud in the 2010s. He believes that after the hype, we’ll find an equilibrium, and AI will just be normal. It’ll be like infrastructure, just like how we no longer think twice about hosting our applications on servers run by AWS or Azure.

AI will become new capability. It’s a broad capability, so let’s break it down.

Superhuman Productivity

I think Grammarly rebranding themselves Superhuman was a smart move. Because that’s what AI unlocks—superhuman productivity in all of us. ChatGPT, Claude, Gemini, and the like help us with research, quick answers, deep dives into unfamiliar subjects, competitive audits, brainstorming, coding, and writing.

In September 2025, OpenAI released a research paper that studied how its users were using ChatGPT:

Patterns of use can also be thought of in terms of Asking, Doing, and Expressing. About half of messages (49%) are “Asking,” a growing and highly rated category that shows people value ChatGPT most as an advisor rather than only for task completion. Doing (40% of usage, including about one third of use for work) encompasses task-oriented interactions such as drafting text, planning, or programming, where the model is enlisted to generate outputs or complete practical work. Expressing (11% of usage) captures uses that are neither asking nor doing, usually involving personal reflection, exploration, and play.

Claude Code is now trending among Silicon Valley firms and its users are not just programmers, but everyone. The use cases include analyzing meeting notes and transcripts, creating daily briefings, and wiring up smart homes. Back in October 2025, Lenny Rachitsky wrote a newsletter highlighting how 50 non-technical people used Claude Code. Non-coders are using it for organizing their desktops, to finding high-quality leads, to cleaning up messy invoice filenames. Agentic tools like this are helping us deal with the drudgery of computing.

That’s general-purpose. For design, it’s similar.

The Superhuman UX Designer

I won’t linger here too long because I’ve covered how AI tools can help our process in other posts and I continue to link to articles and resources on this topic. But, suffice it to say, that AI helps with every step of our design process.

• Discovery: AI-powered user interviews is very nascent, with Anthropic trying it out visibly recently. But using AI to help with synthesis and report writing is a big time-saver. Normal caveats apply: it’s still incumbent on the designer to internalize user research and stand behind the AI-assisted synthesis.
• Design: Using AI to generate some ideas, to visualize and prototype, are all fair game. AI isn’t replacing what we do; it’s another tool to use, like Dribbble once was.
• Develop: Not all designers ship code, but some do. Many web designers build sites for their clients. Today, if I were to create a site for a client, I would much rather prompt my way there in Claude Code than click my way through it in Squarespace.

Using AI tools helps us do our work just a little bit faster. Maybe it unlocks new capabilities, but chances are, we’re just able to fit more work in each day. (Or leave earlier!)

Mission-Critical SaaS Is Here to Stay

Companies depend on mission-critical software to run their businesses. If you’re a designer, are you really going to use Claude Code to make a Figma clone for you so you don’t have to pay them $20 per month? There are a lot of complaints about Adobe’s Creative Cloud prices. Same question, would you spend dozens of hours and thousands of tokens to recreate Photoshop, Lightroom, After Effects, and Illustrator?

The same applies for businesses. A company that makes shoes, like Authentic Brands Group which makes Reeboks, isn’t going to vibe code their CRM because it’s not their core competency. They’d rather pay Salesforce and focus on making great footwear. And same goes for smaller companies, say HVAC, electrical, or plumbing specialty contractors. They will want to use a SaaS product like BuildOps (my employer) to run their businesses so they can focus on keeping America’s infrastructure going.

The argument that building bespoke software for your business is illogical is quite simple once you break down what needs to be done to do that. You’ll have to go through the entire product design and development process, beginning with conducting user research and writing a PRD. Yes, AI can help here, but it’s real work if you want something good. Then you have to go about prompting it to create the UI and the backend. What about authentication and databases? What about security and a mobile app version? And back to our earlier point, what about integrating your bespoke app with other software? What about maintaining all of that?

Spoiler alert: Claude Code is not going to run 24/7 in the background and magically work all that out. There are the real human acts of agency and coordination.

Custom Micro-Apps and Websites

Now here’s the nuance. Small utility applications, let’s call them micro-apps, are easier to vibe code and won’t need much maintenance. If an accounting department gets scans of invoices and receipts and want them renamed a certain way, having Claude Code write a little program to read the scan and rename it is trivial. It can be done in hours before lunch. If there are small utility apps out there, I can see a world where their revenues will drop.

Websites, especially brochure sites, will be relatively easy to vibe code. Casey Newton of Platformer recently canceled his nearly $200 annual Squarespace subscription because he used Claude Code to build his personal website and it’s hosted on Netlify for free.

While the current vibrant ecosystem of visual website builders including Webflow, Framer, and the various WordPress page builders like Elementor and Bricks Builder exists, I can’t see those lasting in the long run. Using an agentic coder to make websites is much less labor intensive. Of course, I don’t believe that Claude Code and the like kill the need for designers. Instead, we still have a place in gathering requirements and compose intention that can be fed into the AI. And whether we start from Figma mockups or dive straight into prompting, that discovery step, that needfinding, and the translation of that to a solution is all necessary before any code is written.

An Opportunity for Web Designers

If there is a market for designers to put together Shopify or Squarespace sites for businesses, there is a market for web designers and developers to put together micro-apps for those same businesses. Again, using Claude Code, Codex, or any AI coding agent, we can make small useful tools for clients. Here are some examples I can imagine:

• An automation that allows a chef-owner to update a daily menu in Google Docs that will then update the menu on their website
• A real estate agent’s listing photo pipeline that automatically watermarks photos with their branding, resizes them for the MLS, Instagram, and their website, and uploads them to the right places
• A boutique’s “new arrivals” emailer where the shop owner adds a photo and price to a spreadsheet, and it auto-generates and sends a styled weekly email to their customer list

There is also a market for designers to create their own products. Sam Altman has said that AI will make it possible for a “one-person billion‑dollar company” to exist. While I personally think that’s a stretch, I do think it’s possible for a designer to now easily create and market their own software products. Every, an amalgamation of media, software, and venture capital, says that each of their core products is essentially run by a single person.

Capability ≠ DIY

Zapier didn’t kill the systems integrators. Squarespace didn’t kill the web designers. AI isn’t going to kill SaaS. The pattern is always the same: a new technology makes something dramatically cheaper and more accessible, pundits declare the incumbents dead, and then the expanded market creates more demand for professional work than existed before. Capability does not equal full-throated DIY. It never has.

I think the enterprise software stocks will recover. The businesses underneath them aren’t going anywhere—not because AI can’t do impressive things, but because running mission-critical software for thousands of companies requires the kind of sustained, coordinated, deeply unsexy work that no one is going to vibe code their way out of. The market is overpricing disruption and underpricing complexity.