Software Ate the World. Now AI Is Eating Software.

In 2011, venture capitalist Marc Andreessen declared “software is eating the world.” Today, AI is beginning to devour software.

Instead of helping the industry improve, new AI tools and agents could end up replacing some software products entirely.

This anxiety crystallized in recent weeks with three developments: Anthropic’s release of a new autonomous AI agent called Cowork, the launch of industry-specific Cowork plugins, and the rise of OpenClaw, an open-source AI assistant spreading rapidly through messaging apps.

“Is this the end for software?” Raimo Lenshow, a veteran tech analyst at Barclays, wrote in a recent note to investors. “The current situation in software feels very unique with the large overhang of how AI will impact the space in the long-run.”

Dual AI threats

Today, companies get stuff done through a complex collection of different software services that help them wrangle data, track financials, sell products, and manage employees, customers, supply chains, and contracts. (If you’ve ever had to login to Workday, you’ll know what I mean).

Generative AI is a threat to these applications in two main ways. First, if employees get more efficient using AI tools, companies may not need to buy as many business software subscriptions. That would dent the growth of “seats,” or how many subscriptions software companies sell. Each employee has a seat, so if there’s no new hiring, growth stalls.

The second threat is more existential. If AI tools and AI agents get good enough, companies could replace the software they use entirely and instead rely on new AI-powered workflows. And with AI coding tools showing big improvements lately, companies could even develop their own software, without needing to buy it from established vendors.

The Anthropic shock

This is why Anthropic’s recent announcements have hit the software sector so hard.

Anthropic’s Cowork marks a clear step beyond chatbots. Instead of simply answering questions, Cowork can plan and execute multi-step tasks on a user’s computer. Users can grant it access to folders, files, and applications, allowing the AI to clean up documents, build spreadsheets and slide decks, analyze data, automate workflows, and even log into web apps to collect information.

Barclays analysts described Cowork as closer to what Microsoft originally envisioned for Copilot—a true digital worker—but with far greater autonomy. Microsoft shares are down about 12% in the past week or so.

Cowork is designed for non-technical and semi-technical users, such as marketers, project managers, and finance professionals, who can direct it using plain English. That matters because it weakens the traditional value proposition of many SaaS tools. If an AI agent can organize files, generate reports, build dashboards, and automate routine workflows on demand, the need for a dedicated, single-purpose application starts to look less obvious.

The threat became more concrete when Anthropic followed up by launching plugins for Cowork this week. These plugins effectively turn the AI into a specialist for roles like sales, finance, legal, marketing, and customer support, connecting it directly to internal data sources and tools. Anthropic even open-sourced a starter set of plugins, signaling an ecosystem approach rather than a closed product.

“This is another front-of-mind example of an AI tool lowering the barrier to entry, gaining traction, and disrupting incumbent workflows,” said Michelle Miller, co-head of the Enterprise Software Technology group at consulting firm AlixPartners.

Why this hits SaaS business models

For years, companies bought software because building it themselves was too slow and expensive. Generative AI is flipping that equation. Tools like Anthropic’s Claude, OpenAI-powered coding assistants, and products from StackBlitz, Replit, and others allow non-engineers to create custom tools with relatively simple English language prompts.

Netlify CEO Matt Biilmann has said his own employees have used AI to build internal replacements for SaaS products like survey and quoting tools. Venture capitalist Martin Casado has described building a personal CRM with AI because it was easier than learning a complex off-the-shelf product. Salesforce, the leading CRM software vendor, is down 42% in the past year.

This shift is especially dangerous for mid-sized SaaS companies. According to AlixPartners, they’re being squeezed between nimble AI-native startups on one side and tech giants bundling AI into existing platforms on the other. Enterprise buyers, under pressure to cut costs, are increasingly asking why they need five tools when one AI tool or agent can do some, or most, of the work.

Pricing adds another layer of tension. AI systems are expensive to run, making traditional per-seat pricing harder to justify. Companies like ServiceNow are experimenting with hybrid and usage-based models. CEO Bill McDermott has insisted AI isn’t hurting results, but investors remain skeptical. ServiceNow shares have plunged 25% in the past month.

The OpenClaw moment

If Cowork represents a top-down push from a well-funded AI lab, OpenClaw, formerly called Moltbot, shows how disruption can also come from the bottom up.

Created as a personal project and released as open source, OpenClaw is a messaging-first AI assistant that works inside platforms like WhatsApp, Slack, and iMessage. Relying heavily on Anthropic’s Claude models, OpenClaw remembers context, proactively nudges users, and can automate tasks using browser actions, scripts, and scheduled jobs. After a month or so of relative obscurity, it went viral in late January.

What caught investors’ attention wasn’t just OpenClaw’s popularity, but what it represents. Instead of logging into multiple apps, users interact with an AI agent in a chat window and let it orchestrate tasks behind the scenes. In effect, the interface becomes conversation, not software menus.

Barclays analysts noted that OpenClaw’s underlying “gateway,” which routes tasks between agents and tools, resembles the orchestration layer that many enterprise software vendors are now racing to build.

The difference is that OpenClaw is free, open source, and user-controlled — a worrying combination for software incumbents that rely on expensive licenses. It’s also a potential problem because the direct user relationship is now controlled by OpenClaw, not software providers.

Eat or be eaten

None of this means software is disappearing overnight. Core systems of record — databases, payroll systems, accounting ledgers — remain deeply embedded in corporate operations. But the layers around them are becoming fluid. Dashboards, workflows, reports, and even entire applications can now be generated and used on the fly by AI agents.

For SaaS companies, the message is stark. Adapt by embracing agents, flexible pricing, and AI-native design — or risk becoming the next thing AI consumes.

“AI is forcing change across software development, AI governance and data security, go-to-market operations, pricing models, valuation frameworks, and business structure,” AlixPartners’ Miller said. “Software companies mastering these transitions will define the winners in the next era, and those unable to adapt will be sidelined as industry fundamentals are redrawn.”

Sign up for BI’s Tech Memo newsletter here. Reach out to me via email at abarr@businessinsider.com.