Wedbush analysts said Tuesday that recent demonstrations of agent-style large language model tools overstate the prospect of replacing existing enterprise software systems. Speaking in the wake of Anthropic's Enterprise Agent event, where the company showcased Claude Cowork in several business contexts, the firm framed these tools as powerful but limited without the surrounding enterprise infrastructure.
At the event, Anthropic presented live examples across industries. Spotify reportedly cut engineering effort for complex code migrations, Novo Nordisk accelerated documentation timelines for clinical studies, and Salesforce shortened processes in Slack. Wedbush emphasized that such gains depend on the quality and accessibility of the data available to the models, and that models alone do not supplant the platforms that manage enterprise workflows.
The analysts cautioned that market observers are conflating improvements in foundation model intelligence with the idea that those models can entirely replace enterprise software. While Anthropic's and OpenAI's demonstrations showcase model reasoning and task handling, Wedbush noted that foundation models lack key enterprise capabilities: workflow orchestration, compliance frameworks, audit trails, security controls, deep integrations, billing and monetization systems, uptime guarantees, and enterprise-grade service level agreements.
Wedbush singled out vendors such as Microsoft (MSFT), Salesforce (CRM), ServiceNow (NOW), and Pegasystems as deeply embedded in enterprise workflow and record-keeping. According to the analysts, displacing these systems would require removing mission-critical infrastructure and the orchestration those platforms provide, not merely adding a large language model to the stack.
Rather than acting as replacements, foundation models are expected to be hosted inside existing workflow engines. The analysts described AI agents as dependent on orchestration layers that coordinate actions across multiple systems - a role that incumbent platforms already hold. This view frames the model layer as a component that can be integrated into workflow engines instead of a substitute for them.
Wedbush also highlighted security and operational consequences. The spread of AI agents and autonomous workflows expands the enterprise attack surface by creating more application programming interfaces, machine identities, opportunities for lateral movement, and cloud-native workloads. Managing this more complex environment will increase demand for run-time monitoring, identity governance, model security, and zero-trust enforcement.
On that basis, the analysts identified CrowdStrike (CRWD), Palo Alto Networks (PANW), and Zscaler (ZS) as likely beneficiaries, calling cyber security the enforcement layer of AI. They further argued that Anthropic and OpenAI do not possess the 20-year enterprise distribution networks, chief information officer relationships, or embedded vertical workflows that incumbent platform vendors have built, and that the model layer may commoditize more quickly than the workflow layer.
Finally, Wedbush suggested that AI's arrival is accelerating deal cycles and prompting a modernization wave rather than bypassing installed software bases. According to the firm, AI lowers friction for legacy transformation projects and is more often a catalyst for modernization than a path to immediate displacement.
Key takeaways
- AI agents demonstrate productivity gains but lack the enterprise infrastructure to replace core platforms.
- Incumbent workflow and systems-of-record vendors retain advantages in orchestration, compliance, and enterprise distribution.
- Security vendors are positioned to benefit as AI increases attack surface and operational complexity.
Impacted sectors: Enterprise software, cybersecurity, cloud services.
Risks and uncertainties
- Security risk - Widespread use of AI agents increases APIs, machine identities, and lateral movement vectors, raising demand for cybersecurity and identity governance.
- Commoditization risk - The model layer may commoditize faster than workflow orchestration, which could pressure vendors focused only on models.
- Integration and compliance uncertainty - AI demonstrations do not address enterprise needs for auditability, billing, and SLAs, leaving questions about practical deployment at scale.