OpenAI just declared war on Salesforce.

I've previously dismissed the idea that AI and vibe coding could threaten incumbent SaaS companies.

Fly-by-night vibe-coded solutions will eventually disappear, and companies attempting to compete will learn why maintaining enterprise software is far more complex and expensive than anticipated.

OpenAI's Frontier made me pause.

This signals that OpenAI is directly challenging the heavyweights (the conspiracy theorist in me wonders if this is why Marc Benioff has been shifting his praise from ChatGPT to Gemini).

I'm already seeing companies fall into the same trap with AI that plagued traditional business software: scaling horizontally. They acquire niche, purpose-built applications to solve specific challenges and fuel growth… until they need to connect these systems and it becomes a nightmare.

Agent orchestration appears to be the answer. One platform to manage permissions, business context, and purpose-built agents across multiple functions. Frontier is a genuine attempt at this unified platform, featuring the hallmarks of enterprise software: identity, permissions, and boundaries are central themes in the release.

"AI is the new UI" feels more real with every announcement from these AI giants.

Let's bring this back to reality: there are significant unknowns. The platform hasn't been publicly tested. AI agents have a long way to go. Prompt injections remain a serious concern. Agents make mistakes constantly. The effectiveness of boundaries and guardrails is questionable.

That said, this technology will improve. And while it does, Salesforce should be careful. Based on the demo videos, I don't think it's coincidental that OpenAI showed Frontier with read/write access to their platform. Agentforce might soon have a rival that can orchestrate agents better than it can.

OpenAI isn't just selling models anymore. They're positioning themselves as the operating system for enterprise AI.