The Shift
The New Developer Is an LLM
Three years ago, a non-technical founder or operations lead who needed to connect two systems hired a developer or waited for IT. Today, that same person opens Claude or Cursor, describes the outcome, and an hour later has a working MCP server or a set of skills that can read invoices, create tickets, update CRM records, and send summary emails.
This is not science fiction. It is Tuesday. The barrier to building production-grade automation has collapsed. The barrier to building production-grade secure automation has not moved at all, and in many cases has gotten worse because the people now doing the building have never carried a pager for a breached integration.
The model is fast, confident, and completely indifferent to your threat model. It will request the broadest OAuth scopes that make the demo work. It will hard-code long-lived API keys. It will create service accounts with directory read and write because the task mentioned "list users." It will store credentials in environment variables that end up in a repo. And it will do all of this while telling you the integration is complete.
You now have a powerful, persistent, often undocumented bridge between an AI that can be tricked or hijacked and the systems that run your business or your clients' businesses. That is the new attack surface. Most teams do not see it until after the first incident.