AI Coding Agents Spark Selloff, but Enterprise Software Holds Its Ground

The idea that tools like Claude Code will replace customer relationship management software sold by firms like Salesforce is “overblown,” Amazon Web Services (AWS) CEO Matt Garman said Tuesday (April 7) at the HumanX conference in San Francisco, according to The Information. AI will be “enormously disruptive,” he said, but also a “huge opportunity.” Software firms that fail to lean in, he warned, “are in trouble.”

Amazon is a shareholder in Anthropic. AWS is a primary cloud provider for Anthropic’s models.

Market Reaction 

Wall Street wiped $285 billion from tech stocks within 24 hours of Anthropic’s February Cowork legal plug-in launch, Fortune reported. When Anthropic launched Claude Code Security on Feb. 20, JFrog dropped 24.61% in a single session, according to Forrester. CrowdStrike, Okta and Zscaler also fell.

Forrester analysts called the damage to identity and runtime detection stocks “sentiment contagion.” Vendors whose products depend on pattern-matching like static application security testing and software composition analysis tools faced direct competitive pressure from AI reasoning capabilities.

Bank of America analyst Vivek Arya called the selloff “overblown” and “logically inconsistent,” according to Fortune. Nvidia CEO Jensen Huang told CNBC the markets “got it wrong.” Most major SaaS companies continue to hit their revenue numbers. Salesforce’s price-to-earnings multiple dropped from roughly 30x to 15x, according to No Jitter.

Enterprise Software Has Deeper Moat

James Cortada, a technology history professor at the University of Minnesota, told Fortune that migrating mission-critical systems using AI-generated code is like “changing flat tires on a car driving at 60 miles an hour.” Enterprise platforms carry years of integrations, compliance infrastructure and operational complexity that AI coding tools do not replicate.

Forrester analysts noted that AI companies do not need to outperform incumbents to compress their margins. Being good enough while bundled into an existing enterprise commitment is sufficient. SaaS spend may not fall in the near term.

No Jitter analyst Dave Michels noted that AI agents require licenses for the tools they use — Salesforce, Slack, Microsoft 365. Deploying agents at scale could increase SaaS expenditure as digital headcount replaces human headcount across workflows.

AI is also shifting pricing. As PYMNTS reported, SaaS is moving toward consumption-based models. Companies are paying for usage and outcomes rather than fixed licenses.

Claude Code’s Performance Issues

Stella Laurenzo, director of the AI group at AMD, filed a GitHub issue last week alleging Claude Code has degraded since February, as reported by The Register. Her team analyzed 6,852 sessions and 234,760 tool calls. Stop-hook violations, indicating shallow reasoning, increased from 0 to 10 per day starting March 8.

Code reads per session dropped from 6.6 to 2. Laurenzo’s team has switched to another provider and asked Anthropic to introduce a high-thinking subscription tier for complex engineering work.

Security risks are emerging as well. Hackers are exploiting leaked Claude Code configurations to distribute malware. The incidents highlight how quickly vulnerabilities can spread when AI agents gain access to development environments and enterprise workflows, according to TechRadar.

Ultimately, the question is not whether AI is killing software. It is how it is changing it.

Rick Sherlund, senior adviser at Wedbush Securities, said in a Goldman Sachs research note that the industry is being rebuilt around AI rather than replaced.

“AI is expressed in software, so declaring software ‘dead’ is misguided,” he said. “Software will endure, even if the industry looks different. Some segments will see disruption. The broader industry will adapt and benefit as AI spreads across business functions.”

For all PYMNTS AI and digital transformation coverage, subscribe to the daily AI and Digital Transformation Newsletters.