Wedbush attended Anthropic's Enterprise Agent event on 24 February and came away with a contrarian read: the competitive threat to established enterprise software is overstated, and the companies most likely to benefit from accelerating AI adoption are the platforms investors have been selling down on disruption fears.
The event featured live demos from senior leadership and use cases spanning multiple industries. Spotify used Claude to reduce engineering time on complex code migrations. Novo Nordisk improved clinical study documentation timelines. Salesforce cut turnaround times inside Slack. Wedbush called the demonstrations impressive, then immediately qualified them: the new AI tools will not rip and replace existing software ecosystems, and are only as useful as the data they can reach.
Why foundation models are not enterprise software
The core argument from Wedbush is that the market is conflating two different things. A foundation model demonstrates raw intelligence. An enterprise software platform provides workflow orchestration, compliance infrastructure, auditability, security controls, system integrations, billing, uptime guarantees and enterprise-grade service level agreements. Anthropic and OpenAI are showing the first. Companies like Microsoft, Salesforce, ServiceNow and Pegasystems have spent decades building the second.
Replacing a system of record does not mean layering in a language model. It means ripping out mission-critical infrastructure, and enterprises do not do that quickly or casually. The more relevant dynamic, Wedbush argues, is that AI agents need orchestration layers to execute actions across systems, and the platforms investors fear are being displaced already own that orchestration. Foundation models do not replace workflow engines; they run inside them.
AI as a complexity multiplier for cybersecurity
Wedbush pushes back equally hard on the idea that AI threatens cybersecurity spending. The logic runs in the opposite direction. Deploying AI agents and autonomous workflows expands the attack surface: more APIs, more machine identities, more lateral movement risk, more cloud-native workloads requiring protection. The more enterprises deploy language model-powered agents, the more run-time monitoring, identity governance, model security and zero-trust enforcement they require.
The firm frames cybersecurity as the enforcement layer of AI rather than a casualty of it, and identifies CrowdStrike, Palo Alto Networks and Zscaler as winners in this environment.
Distribution as the durable moat
The third leg of Wedbush's argument concerns enterprise distribution. Anthropic and OpenAI do not have 20-year relationships with chief information officers, embedded vertical workflows, or the kind of account penetration that makes displacement difficult even when a competing product is technically superior. Microsoft, Salesforce and ServiceNow sit at the application layer where business logic lives, and the model layer is likely to commoditise faster than the workflow layer above it.
Platform owners historically capture disproportionate value when new infrastructure emerges, because the new infrastructure increases the strategic importance of the layer that orchestrates it. Wedbush concludes that AI urgency is accelerating deal cycles and lowering the friction of legacy transformation projects, making a modernisation cycle that strengthens the installed base more probable than a bypass that routes around it.
