The emergence of artificial intelligence (AI) coding agents has once again tested the perceived fragility of enterprise software, triggering a significant market reaction. Within 24 hours of Anthropic’s February Cowork legal plug-in launch, Wall Street wiped an estimated $285 billion from tech stocks, as reported by Fortune. This selloff, however, is being met with skepticism by some industry veterans and analysts who assert that enterprise software, much like its resilience against past threats from cloud computing and mobile, possesses a deeper moat than current AI tools can breach.
AWS CEO Matt Garman, speaking on Tuesday (April 7) at the HumanX conference in San Francisco, dismissed the notion that tools like Claude Code would replace established customer relationship management software from firms such as Salesforce as ‘overblown,’ according to The Information. While acknowledging AI’s potential to be ‘enormously disruptive’ and a ‘huge opportunity,’ Garman, whose company Amazon is a shareholder in Anthropic and a primary cloud provider for its models, warned that software firms failing to ‘lean in’ would find themselves ‘in trouble.’
Market Volatility and Analyst Rebuttals
The market’s apprehension was palpable following Anthropic’s product launches. When Claude Code Security debuted on February 20, JFrog experienced a sharp 24.61% drop in a single trading session, as documented by Forrester. Other prominent cybersecurity and identity management firms, including CrowdStrike, Okta, and Zscaler, also saw their stock prices fall. Forrester analysts characterized this impact on identity and runtime detection stocks as ‘sentiment contagion,’ noting that vendors relying on pattern-matching, such such as static application security testing and software composition analysis tools, faced direct competitive pressure from AI’s reasoning capabilities.
Despite the immediate market reaction, several financial experts quickly labeled the selloff as excessive. Bank of America analyst Vivek Arya described it as ‘overblown’ and ‘logically inconsistent,’ according to Fortune. Nvidia CEO Jensen Huang echoed this sentiment, telling CNBC that the markets ‘got it wrong.’ Indeed, most major SaaS companies have continued to meet their revenue targets. While Salesforce’s price-to-earnings multiple did drop from approximately 30x to 15x, according to No Jitter, this shift reflects a re-evaluation rather than an immediate collapse in business fundamentals.
Enterprise Software’s Deep Moat
The enduring strength of enterprise software lies in its inherent complexity and deep integration into mission-critical business operations. James Cortada, a technology history professor at the University of Minnesota, articulated this to Fortune, likening the migration of such systems using AI-generated code to ‘changing flat tires on a car driving at 60 miles an hour.’ Enterprise platforms are built upon years of intricate integrations, robust compliance infrastructure, and operational complexities that current AI coding tools are not equipped to replicate comprehensively.
Forrester analysts further observed that AI companies do not necessarily need to outperform incumbents to exert pressure; merely being ‘good enough’ while bundled into an existing enterprise commitment can suffice to compress margins. However, this does not automatically translate to a near-term reduction in overall SaaS expenditure. Dave Michels, an analyst at No Jitter, pointed out that AI agents themselves require licenses for the tools they utilize, such as Salesforce, Slack, and Microsoft 365. Consequently, deploying AI agents at scale could, paradoxically, increase SaaS expenditure as digital headcount supplements or replaces human headcount across various workflows. Furthermore, AI is actively reshaping pricing models, with SaaS increasingly shifting toward consumption-based models where companies pay for usage and outcomes rather than fixed licenses, as reported by PYMNTS.
Performance and Security Concerns Emerge
Beyond the strategic arguments, practical limitations and risks of current AI coding agents are becoming apparent. Stella Laurenzo, director of the AI group at AMD, recently filed a GitHub issue alleging a degradation in Claude Code’s performance since February, as reported by The Register. Her team’s analysis of 6,852 sessions and 234,760 tool calls revealed a concerning trend: stop-hook violations, indicative of shallow reasoning, increased from 0 to 10 per day starting March 8. Concurrently, code reads per session dropped significantly from 6.6 to 2. Laurenzo’s team has since switched to another provider and has requested Anthropic introduce a ‘high-thinking’ subscription tier for complex engineering work.
Security risks are also emerging as a critical concern. TechRadar reported incidents where hackers are exploiting leaked Claude Code configurations to distribute malware. These events underscore how rapidly vulnerabilities can propagate when AI agents gain access to development environments and enterprise workflows, posing new challenges for cybersecurity.
Ultimately, the prevailing view among many industry observers is that AI is not poised to kill software but rather to fundamentally change it. Rick Sherlund, a senior adviser at Wedbush Securities, noted in a Goldman Sachs research note that the industry is being ‘rebuilt around AI rather than replaced.’ He asserted, ‘AI is expressed in software, so declaring software ‘dead’ is misguided.’ Sherlund concluded that ‘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,’ signaling a period of transformation rather than outright obsolescence for the enterprise software sector.
